JP2003157243A - Device and method for parallel image processing - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a parallel image processor for efficiently carrying out an algorithm for image processing in which processing load varys. SOLUTION: The parallel image processor is provided with a image area assignment means 2 and a processing load calculation means 8. The image area assignment means 2 assigns a processing area and a processing restriction quantity to respective processors and when reassignment is requested from a processor, the means 2 reassigns an area whose assignment is requested as an unprocessed area. The processing load calculation means 8 calculates the processing load in an object image area about the assigned processing area by the respective processors, and instructs to process it a corresponding image processing execution means 9 about a range which does not exceed a designated processing restricted quantity. In the case of exceeding the processing restricted quantity, the means 8 requests reassignment of the area of restriction over to the means 2.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a parallel image processing apparatus and a parallel image processing method for efficiently executing an image processing algorithm whose processing load changes depending on an input image on a parallel computer.

[0002]

2. Description of the Related Art Some algorithms for image processing change the processing load depending on the input image. One of such algorithms is MUSIC (Multiple
There is a Signal Classification). MUSIC is a SAR (Synthetic Ap)
image radar: Synthetic aperture radar) Used for improving image resolution. When using MUSIC to improve the resolution of a SAR image, first, for each row (or column, for simplicity of description, only the row direction will be shown below),
Estimate the number of scattering points based on the eigenvalues. Next, the resolution improvement processing of the SAR image is performed based on the estimated number of scattering points. Therefore, the calculation amount of each row depends on the estimated number of scattering points. Since this number of scattering points depends on the input image,
If the input image is not processed, the processing load of each line is unknown. On the other hand, one of the techniques for speeding up image processing is parallel processing using a plurality of processors. Parallel processing is effective in cases where the image processing size is large and the processing load is high, since there are few dependencies within the image processing. In order to efficiently perform parallel processing, it is important to evenly allocate the load to each processor and to efficiently perform the processing in each processor. In MUSIC, the processing load of each line differs depending on the input image. If such an image processing algorithm is simply parallelized based on the size of the image area (divided based on the size of the area), the load may not be evenly allocated to each processor. If the load cannot be evenly allocated to the processors, the execution efficiency of parallel processing is reduced. As described above, in the parallelization of the algorithm in which the processing load changes depending on the input image, the execution efficiency may decrease with a simple method.

As a general method of dealing with this,
Dynamic scheduling (data division) and calculation of scattering points in advance can be considered. As dynamic scheduling to deal with this problem, depending on the load at runtime,
A method of determining the division of data can be considered. One of such methods is Guided-Self scheduling. In the Guided-Self scheduling, each processor dynamically requests a processing area allocation.
Further, the allocation area is dynamically set (gradually decreased) (see FIG. 7). In general, this scheduling method can suppress variations in the number of allocation requests and divided areas in many cases. Therefore, it is known as an efficient dynamic scheduling method. When the MUSIC is parallelized by the Guided-Self scheduling, the processor that has processed a row including a large number of scattering points can absorb the variation in the processing load to some extent because the number of times the processing area is allocated decreases. However, it does not work when the processing load is extremely high on a particular row. For example, there is a case where the load is high in the last one row, or a case where only a specific row takes a processing time which is several times the image size or more. In order to solve this, it is necessary to change the method of dividing and allocating the image area (problem). To change the division method, it is necessary to consider the characteristics of the reference information (the number of scattering points in MUSIC). However, there is no general solution to this problem to get the optimal solution. Also, there is no conventional example. on the other hand,
The pre-calculation method has the following problems. MUSIC
When the resolution of the SAR image is improved by each line by using, the cache data (data already stored in the cache memory of the processor) used in the calculation of the scattered point number is calculated in the resolution of the next calculation phase. There is an advantage that it can also be used in improvement processing. When the scatter point estimation of all rows is executed in advance, the cache data used in the calculation of scatter point estimation in the resolution improvement processing phase cannot be used.
Therefore, there is a problem that the efficiency of the processing executed by each processor is reduced. In addition, a memory area for counting the scattering points is newly required. In addition, here, MUSIC
However, similar to MUSIC, the same problem occurs in the image processing algorithm in which the processing load changes depending on the input image.

[0004]

Since the conventional parallel image processing method is configured as described above, when adopting the dynamic scheduling, it is necessary to change the method of dividing and allocating the image area. In order to change the division method, it is necessary to consider the characteristics of the reference information (scattering point number in MUSIC), and there is a problem that there is no general solution for obtaining this optimal solution. In addition, the method of calculating the scattered points in advance has a problem that the efficiency of the processing executed by each processor decreases.

The present invention has been made in order to solve the above problems, and is an element (MUSI) serving as a load reference.
It is possible to obtain a parallel image processing device and a parallel image processing method for efficiently executing an algorithm for image processing in which the processing load changes depending on the input image, by reallocating the processing area based on the scattering point number in the case of C). To aim.

Further, according to the present invention, the size of the area to be allocated is adjusted in response to a request for reallocation of the processing area to make the allocation of the processing areas to the respective processors more uniform, thereby making it possible to perform the image processing. It is also an object to obtain a parallel image processing device and a parallel image processing method that efficiently execute an algorithm.

Furthermore, the present invention allocates a processing area to each processor in consideration of a range in which the processing allocated to each computer can be efficiently executed, such as a range in which the cache can be efficiently accessed. An object of the present invention is to obtain a parallel image processing device and a parallel image processing method for executing algorithms in parallel more efficiently.

Further, according to the present invention, after allocating an image area,
The calculation load is significantly increased on a certain processor, and reallocation in the minimum image area (one row area in the case of MUSIC) for calculating the processing load calculation and image processing execution set in the image processing algorithm is sufficient. A parallel image processing apparatus and a parallel image processing apparatus that perform parallel execution more efficiently by dividing the image area of the smallest unit (in the case of MUSIC, dividing the processing for one row) and executing it as a reallocation target when it is not possible The purpose is to obtain an image processing method.

Further, according to the present invention, when the processing is distributed and executed by the dynamic scheduling, it is difficult to evenly finish the processing in each processor when a high load portion appears in the latter half of the processing. It is disadvantageous in parallel execution,
Using the information about the input image that can be obtained by simple calculation, predict the load of the algorithm for image processing of the target, and use the prediction result to transfer from the processing area predicted to be heavy to each processor It is an object of the present invention to obtain a parallel image processing device and a parallel image processing method for efficiently executing an algorithm for image processing of a target by allocating the.

Further, the present invention provides a parallel image processing apparatus and a parallel image processing method for efficiently executing an algorithm for image processing of a target by predicting a load by utilizing a condition of capturing an image. To aim.

Further, according to the present invention, when the difference between the maximum processing time and the minimum processing time in the algorithm for the target image processing is large enough to make efficient parallel execution difficult by dynamic reallocation of the processing area. By detecting this, calculating the load of the entire image area in advance, and determining the processing area allocation to each processor based on the calculation result, the algorithm for the target image processing can be set even under difficult conditions. An object is to obtain a parallel image processing device and a parallel image processing method that execute efficiently.

[0012]

A parallel image processing apparatus according to the present invention comprises an area division method setting means for setting a processing area allocation method according to the number of processors used and an image processing area size, and a maximum value for a unit area. A processing limit amount calculation unit that sets a processing limit amount according to the processing load amount and the calculated processing region, and a processing region is calculated based on the allocation method set by the region division method setting unit, and the processing region and its While assigning the processing limit amount set by the processing limit amount calculation means to each processor according to the processing area,
When the processor requests re-allocation, the image area allocating means for re-allocating the area requested for re-allocation as an unprocessed area, and the processing area allocated by the image area allocating means in each processor. hand,
The processing load in the target image area is calculated, the processing is instructed to the corresponding image processing execution means in the range not exceeding the specified processing limit amount, and when the processing limit amount is exceeded, the limit overrange area is reached. And a processing load calculating means for requesting the image area allocating means to reallocate.

In the parallel image processing apparatus according to the present invention, the area division method setting means sets the processing area allocation method according to the maximum processing load amount in the unit area set in the maximum processing amount setting means per unit area. The load calculation type area division method setting means is set.

The parallel image processing apparatus according to the present invention comprises the area division method setting means, the processing load measuring means for actually measuring the execution time for the maximum load and the execution time for the minimum load for the unit area in the image processing algorithm, The area division method setting means is a measurement corresponding area division method setting means for setting the processing area allocation method according to the actual measurement time measured by the processing load measuring means, and the processing limit calculation means is
This is a measurement corresponding processing limit amount calculating means for setting the processing limit amount according to the actual measurement time measured by the processing load measuring means.

The parallel image processing apparatus according to the present invention comprises processing block setting means for setting an image area of an arbitrary size as a block, and the processing load calculation means preferentially processes the block set by the processing block setting means. The processing block-corresponding processing load calculating means is used, and the image processing executing means is processing block-corresponding image processing executing means for executing the image processing of the image area designated by the processing block-corresponding processing load calculating means.

The parallel image processing apparatus according to the present invention is set by the cache information setting means for setting the cache information in the target computer, the image processing area size set by the processed image information setting means and the cache information setting means. And processing block calculation means for setting the size of the block set by the processing block setting means in accordance with the cache information.

In the parallel image processing apparatus according to the present invention, the processing block setting means sets the processing area calculated by the image area allocating means and the size of the block designated by the processing block corresponding processing load calculating means. It is provided with a dynamic processing block calculating means for dynamically changing and setting the size of the block to be processed.

In the parallel image processing apparatus according to the present invention, the area division method setting means sets the processing area allocation method for setting the processing area allocation method according to the size of the block set by the processing block setting means. It is a means.

In the parallel image processing apparatus according to the present invention, the processing limit amount calculating means is a processing block corresponding type processing limit amount calculating means for setting the processing limit amount according to the size of the block set by the processing block setting means. It was done.

In the parallel image processing apparatus according to the present invention, the area division setting means is a unit area division corresponding area division setting means for setting a processing area allocation method in consideration of division of the unit area, and the processing limit amount calculating means Is a unit area division-compatible processing limit amount calculation means for setting a processing limitation amount when the unit area is divided, and the image area allocating means divides the unit area to allocate and reallocate the processing areas. The image processing execution means is a division corresponding image area allocating means, the processing load calculating means is a unit area division corresponding processing load calculating means for judging a processing load calculation and presence / absence of reallocation corresponding to division of a unit area. Is a unit area division corresponding image processing executing means for executing image processing of an image area designated corresponding to division of a unit area.

The parallel image processing apparatus according to the present invention refers to the information of the unit area division correspondence type area division setting means to judge whether or not division of the unit area is possible, and transmits it to the unit area division correspondence type processing load calculation means. The unit area division permission / prohibition condition setting means is provided.

In the parallel image processing apparatus according to the present invention, when the unit area is divided and reallocated, the calculation result of the unit-area-division-compatible processing load calculating unit which first calculates the processing load in the unit area is displayed. It is provided with a pre-calculation result holding unit that holds the divided unit area and allows the unit area division-compatible processing load calculation means of another processor to which the reassigned unit area is referred to the calculation result.

A parallel image processing apparatus according to the present invention is a cache information setting means for setting cache information in a target computer, and cache information set in the cache information setting means and image processing set in the processed image information setting means. A unit area division cost calculation means for calculating a cost when the unit area is divided according to the area size, and for calculating a processing limit amount from the calculated cost and transmitting it to the unit area division correspondence processing load calculation means; It is equipped with.

The parallel image processing apparatus according to the present invention comprises a load prediction setting means for setting a prediction load for each unit area in the image processing algorithm, and the area division method means is used for the prediction set by the load prediction setting means. This is a load prediction type area division method means for setting a processing area allocation method according to a load.

A parallel image processing apparatus according to the present invention is a brightness value calculation formula setting means for setting a calculation formula of a brightness value of a target image, and a target image according to the calculation formula set by the brightness value calculation formula setting means. A brightness value calculation executing means for calculating a brightness value for each pixel, and a pixel reference load prediction calculating means for calculating a predicted load according to the brightness value calculated by the brightness value calculation executing means and transmitting it to the load prediction setting means. Be prepared.

The parallel image processing apparatus according to the present invention comprises pixel reference load prediction condition setting means for setting the calculation method of the prediction load for the brightness value of each pixel, and the pixel reference load prediction calculation means is used for the pixel reference load prediction condition. This is a measurement condition corresponding pixel reference load prediction calculation means for calculating the prediction load for each unit area according to the calculation method set in the setting means.

In the parallel image processing apparatus according to the present invention, a characteristic region extraction setting means for setting a method of calculating a characteristic region serving as a reference of load prediction from the brightness value of a pixel, and a calculation method set by the characteristic region extraction setting means And a characteristic region extracting unit that extracts a characteristic region from the luminance value of each pixel of the target image calculated by the luminance value calculation executing unit according to It is a characteristic region load prediction calculation means for calculating a predicted load for each unit area in the image processing algorithm according to the characteristic region and transmitting it to the load prediction setting means.

The parallel image processing apparatus according to the present invention comprises a characteristic region load prediction condition setting means for setting a method of calculating a predicted load for the characteristic region, and the characteristic region load prediction calculation means is operated by the characteristic region load prediction condition setting means. This is a characteristic condition load predictive calculation means corresponding to a measurement condition for calculating a predicted load for each unit area according to the set calculation method.

The parallel image processing apparatus according to the present invention comprises a reference area setting means for setting an image area in a fixed range as a reference area, and the pixel reference load prediction calculation means is set within the reference area set by the reference area setting means. The reference area load prediction calculation means calculates the predicted load for each unit area in the image processing algorithm according to the brightness value calculated by the brightness value calculation execution means.

The parallel image processing apparatus according to the present invention comprises the reference area load prediction condition setting means for setting the calculation method of the prediction load in accordance with the brightness value in the reference area, and the reference area load prediction calculation means is used as the reference area. This is a measurement condition-corresponding reference area load prediction calculation means for calculating a predicted load for each unit area according to the calculation method set by the load prediction condition setting means.

In the parallel image processing apparatus according to the present invention, the inclination reference setting means for setting the inclination of the reference luminance value and the pixel reference load prediction calculation means are set to the inclination of the luminance value set by the inclination reference setting means. According to this, the gradient load prediction calculation means calculates the predicted load for each unit area.

The parallel image processing apparatus according to the present invention is provided with the gradient load prediction condition setting means for setting the calculation method of the prediction load with respect to the gradient of the brightness value, and the gradient load prediction calculation means is provided.
The tilt load prediction condition setting means is a tilt load prediction calculation means corresponding to the measurement condition for calculating the prediction load for each unit area according to the calculation method set by the tilt load prediction condition setting means.

The parallel image processing apparatus according to the present invention comprises a photographing area load prediction calculating means for calculating the predicted load for each unit area in the image processing algorithm from the area information of the image and transmitting it to the load prediction setting means. Be prepared.

The parallel image processing apparatus according to the present invention comprises a brightness value calculation formula setting means, a brightness value calculation executing means, a pixel reference load prediction condition setting means, a measurement condition corresponding pixel reference load prediction calculation means, and a characteristic region extraction. Setting means and characteristic area extraction means, characteristic area load prediction condition setting means and measurement condition corresponding characteristic area load prediction calculation means, reference area setting means, reference area load prediction condition setting means and measurement condition corresponding reference area load Prediction calculation means, inclination reference setting means,
The tilt load prediction condition setting means and the measurement condition-compatible tilt load prediction calculation means, the photographing area load prediction calculation means, and the prediction calculation weighting setting means capable of weighting the prediction conditions are provided.

The parallel image processing apparatus according to the present invention comprises a prediction value recording means for recording the calculated prediction value and the input condition, and a processing result recording means for recording the processing load calculated in actual execution for the prediction value. And a prediction parameter correction means for correcting the calculation method of the predicted load based on the predicted value recorded by the predicted value recording means and the processing load recorded by the processing result recording means.

The parallel image processing apparatus according to the present invention records the conditions under which the images at the time of executing the past prediction calculation are captured, and at the same time, the optimum prediction calculation is performed based on the imaging conditions for the image for which new prediction calculation is performed. It is provided with a photographing condition load prediction correction means for selecting correction.

The parallel image processing apparatus according to the present invention, when instructed to calculate a new predicted value, searches for and selects a similar condition among the input conditions for calculating the predicted value recorded in the predicted value recording means. At the same time, the optimum prediction condition calculation means for correcting the selected prediction calculation with reference to the calculation result recorded in the processing result recording means is provided.

The parallel image processing apparatus according to the present invention comprises a pre-calculation selection setting means for setting a condition as to whether or not to perform a pre-calculation processing, and a pre-calculation processing based on the condition set by the pre-calculation selection setting means. A pre-calculation determining unit that determines the presence / absence and a pre-processing load amount that pre-executes a calculation for obtaining a processing load amount for each unit region for the entire image region when instructed to perform the process by the pre-calculation determining unit. Based on the calculation load and the processing load amount for each unit area pre-calculated by the pre-processing load amount calculating means,
It is provided with a pre-processing compatible image area allocating means for allocating areas so that the processing loads of the respective processors become equal.

In the parallel image processing apparatus according to the present invention, the area division setting means is the image information reference type area division setting means for setting the processing area allocation method in consideration of the image information, and the image area allocation means from the processor. When the reallocation is requested, the image information reference type image area allocating means for reallocating the processing area by referring to the image information is used, and the processing load calculating means is provided for each processing unit with respect to the allocated processing area. This is an image information reference type processing load calculation means for calculating image information.

In the parallel image processing apparatus according to the present invention, in the image information reference type image area allocating means, when reallocating the processing areas by referring to the image information, the reallocation is performed by giving priority to the number of unit areas. The processing area is determined.

In the parallel image processing apparatus according to the present invention, in the image information reference type image area allocating means, when the processing areas are reallocated by referring to the image information, the size of the image information is preferentially reallocated. The processing area is determined.

In the parallel image processing apparatus according to the present invention, in the image information reference type image area allocating means, when the processing area is re-allocated by referring to the image information, the processing area is simply re-allocated from the image information. Instead of adopting, the processing load of the image processing to be executed is calculated based on the obtained image information, and the processing area to be reallocated is determined based on the processing load.

In the parallel image processing apparatus according to the present invention, in the image information reference type processing load calculating means, at the time of initial allocation,
The calculation of the image information and the instruction of the processing to the image processing execution means are continuously executed.

According to the parallel image processing method of the present invention, when a parallel computer executes an image processing algorithm whose processing load changes depending on an input image, the allocation processing area is dynamically determined and the processing for the allocation processing area is performed. A limit amount is calculated from the maximum processing load amount in a unit area, and is assigned to the algorithm execution means of each processor as a set of an assigned processing area and a process limit amount, and the area exceeding the process limit amount from the algorithm execution means of each processor Is received, the area is incorporated into the entire unprocessed area, and is again assigned to the algorithm execution means of each processor as the processing area. In each processor, the processing load in the target image area is assigned to the assigned image area. Calculate and instruct execution of processing in the range that does not exceed the specified processing limit, If it exceeds the physical limit amount is obtained so as to request the re-allocation of space limitations over.

In the parallel image processing method according to the present invention, the allocation method of the allocation processing area is determined from the maximum processing load amount in the unit area in the image processing algorithm.

In the parallel image processing method according to the present invention, the execution time with respect to the maximum load and the execution time with respect to the minimum load in the unit area of the image processing algorithm are measured, and the allocation method and the processing restriction of the allocation processing area are measured from the measured time. The method of calculating the quantity is decided.

In the parallel image processing method according to the present invention, an image area of an arbitrary size is designated as a block, and the designated processing is preferentially performed within the block.

In the parallel image processing method according to the present invention, when the processing load on a certain processor is remarkably increased and the reallocation based on the unit area in the image processing algorithm is not sufficient, the unit area is divided and reallocated. In addition to performing the allocation, each processor also determines the specified processing restriction amount in consideration of division of the unit area and executes the processing in the divided unit area.

In the parallel image processing method according to the present invention, whether or not the unit area can be divided is determined based on the number of waiting processors and the amount of the remaining processing area.

In the parallel image processing method according to the present invention, when the unit area is divided and the reallocation is requested, the processor which processes the unit area first executes the calculation result of the first half portion for obtaining the processing load. Is recorded, and the remaining processing in the divided unit area is performed by the processor instructed later, using the calculation result of the recorded first half part to perform the second half part processing. Is.

In the parallel image processing method according to the present invention, when the unit area is divided, the division position where the cache access is advantageous is obtained from the sizes of the cache and the image area.

The parallel image processing method according to the present invention sets the prediction load for each unit area in the image processing algorithm,
The image area allocation method is determined based on the predicted load.

In the parallel image processing method according to the present invention, the prediction load is calculated based on the brightness value of each pixel of the image to be processed.

In the parallel image processing method according to the present invention, the characteristic region is determined from the brightness value of each pixel of the image to be processed, and the prediction load is calculated based on the characteristic region.

In the parallel image processing method according to the present invention, an image area in a fixed range is set as a reference area, and the prediction load is calculated by using the brightness value of each pixel in the reference area. .

The parallel image processing method according to the present invention records the calculated predicted value and the input condition for the predicted value, and
The processing load calculated by actual execution is recorded for the predicted value, and the calculation method of the predicted load is corrected based on these recorded values.

The parallel image processing method according to the present invention is a method for newly calculating a predicted value from the image capturing condition of the past prediction calculation execution and the input condition used as the calculation standard of the predicted value. The selection is made and an appropriate correction is made.

The parallel image processing method according to the present invention is executed by determining in advance whether or not the processing load for each unit area is obtained, and when the processing load for each unit area is obtained in advance. The image area allocation method is determined based on the load.

[0059]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below. Embodiment 1. 1 is a block diagram showing a parallel image processing apparatus according to Embodiment 1 of the present invention. In the figure, 1 is a parallel image processing apparatus, 2 is an image area allocating means, and 3 is a maximum processing amount setting means per unit area. Reference numeral 4 is a processor number setting means, 5 is a processed image information setting means, 6 is a region division method setting means, 7 is a processing limit amount calculating means, 8 is a processing load calculating means, 9 is an image processing executing means, and 10 is a platform. ,
Reference numeral 11 is an image processing program, 12 is a processing part for determining the processing load, and 13 is a processing part for dynamically changing the load. 2 is a flow chart showing the processing of the image area allocating means, FIG. 3 is a flow chart showing the processing of the processing load calculating means, FIG. 4 is a block diagram showing the mounting image of the parallel image processing device at the time of parallel execution, and FIG. 5 is MUSIC. FIG. 7 is an explanatory view showing an outline of the operation of a program for executing the super-resolution processing of the SAR image used, FIG. 6 is an explanatory view showing the concept of the operation of the execution management table, and FIG. 7 is an allocation of the processing area in the guided-self scheduling. 9 is an explanatory diagram showing an example, FIG. 8 is an explanatory diagram showing examples of a “super-resolution execution region” and an “unprocessed region” for the “next processing region” assigned to the processing load calculation means, FIG. 9
Is an explanatory view showing the estimated number of scattering points and processing load (processing time) for a certain sample image in the super-resolution processing of the SAR image using MUSIC which is the processing target, and FIG. 10 is an implementation when processing with a 4-processor configuration It is explanatory drawing which shows an image.

Next, the operation will be described. The platform 10 is a platform for a parallel computer H / W, an OS, a parallelization library, and the like. Image processing program 11
Is a program for executing super-resolution processing of SAR image using MUSIC. As shown in FIG. 5, the image processing program 11 advances the processing line by line. The process in each row can be divided into a part for performing the process of estimating the number of scattering points based on the eigenvalue and a part for performing the resolution improving process of the SAR image based on the estimated number of scattering points. The processing part 12 for determining the processing load is a part of the image processing program 11 for performing processing for estimating the number of scattering points based on the eigenvalue for each row. The time required to estimate the number of scattering points in this process is the same (constant) in each row. In the image processing program 11, the processing part 13 whose load dynamically changes is
This is a part for performing the resolution improvement processing of the SAR image based on the estimated number of scattering points. The time required for this processing is proportional to the estimated number of scattering points.

The parallel image processing apparatus 1 comprises a platform 1
0, and in the super-resolution processing of the SAR image using MUSIC, an area to execute the processing is assigned to each processor, and the assigned area is assigned to the image processing program 11
It is a device to be executed by. As a method of implementing the parallel image processing apparatus 1, there are a method of independently providing as a parallelization support library, a method of incorporating into a platform OS or a parallelization library, and a method of incorporating into an image processing program. In the first embodiment, a method of independently providing the parallelization support library will be described.

When the parallel image processing apparatus 1 executes in parallel, as shown in FIG. 4, the processing load calculating means 8 and the image processing executing means 9 are arranged one for each processor. In the example of FIG. 4, if the number of processors is N, the processing load calculation means 8
-1, processing load calculation means 8-2, ..., Processing load calculation means 8-N, one for each processor, total N.
There are individual processing load calculation means 8 and image processing execution means 9.
In the first embodiment, the shared memory type SMP (Symmetric M) is used as the H / W configuration of the platform 10.
ultiprocessor). Here, there are means implemented as a common part among the processors and means implemented by being divided into each processor. In the first embodiment, the image area allocating means 2, the processing limit amount calculating means 7, the image processing program 11 and the like are realized as common parts among the processors, and the processing load calculating means 8 and the image processing executing means 9 are implemented. Will be described as being implemented as a means realized by being divided into each processor.

The operation of the image area allocating means 2 will be described with reference to FIG. The image area allocation unit 2 first obtains information necessary for initialization from the maximum processing amount setting unit 3 per unit area, the used processor number setting unit 4, and the region division method setting unit 6. From the used processor number setting means 4,
The maximum number of processors to be used is obtained from the maximum processing amount setting unit 3 per unit area. In the example of the super-resolution processing of the SAR image using MUSIC, the unit area is a row, and the maximum processing number is the maximum value of the number of scattering points estimated in each row. From the area division method setting means 6, the size of the entire image processing area and the allocation method of the processing areas are obtained. The area division method setting means 6 is the processing image information setting means 5
To image processing area information (image size, pixel size,
The number of processors to be used is acquired from the data type, etc.) and the number of processors to be used setting unit 4. Area division method setting means 6
Then, the size of the image is obtained from the processed image information setting means 5 and the number of processors to be used is obtained from the used processor number setting means 4, and the allocation method of the processing area is determined. This calculation method is
It can be set arbitrarily. In this example, the processing load is assumed to be uniform in each row, and the processing area allocation method is based on the general dynamic area partitioning method of parallel processing. However, in this apparatus, a reallocation request is made, and an image area to be processed may increase in the middle. Therefore, the area division method setting means 6
Then, the allocation method is selected on the assumption that the image area is increased by the reallocation request. Next, the image area allocating means 2 initializes internal information using the obtained information. Here, the "number of remaining processing areas" is calculated from the "size of the entire image processing area" and the number of processing units (in this example,
Number of lines) and set the running flags of all processors to "O"
Set to N ". Furthermore, an execution management table for each processing unit is created from the "size of the entire image processing area", the management flags for all processing units are set to "not", and the head of the "image allocation pointer" is set to the head of the image area. Set (see FIG. 6). Further, the processing load calculation means 8 of all the processors are instructed to start the execution (step ST1). After completing the initialization up to this point, the image area allocating means 2 is in a “waiting for request” state from the processing load calculating means 8 of each processor (step ST2).

The operation when the "next processing" is requested from the processing load calculating means 8 in the "waiting for request" state will be described. First, the "number of remaining processing areas" is checked (step ST3). If the "number of remaining processing areas" is greater than 0, the next processing area allocation operation is performed. In the processing area allocation operation, first, the "next processing area" to be allocated to the processing load calculation means 8 of the request source is determined (step ST4). The size of the “next processing area” to be assigned here follows the allocation method of the processing area designated by the area division method setting means 6 (in this example, the method of calculating the size of the “next processing area”). From the position indicated by the "image allocation pointer", sequentially select the areas for which the management flag is "not" for "the size of the next processing area",
Select it as the "next processing area". When the end of the processing area is reached during selection, the area is returned to the beginning and the areas having the management flag of "not yet" are sequentially selected. When the "next processing area" is determined, the management flag of the execution management table for each processing unit corresponding to the area is set to "total" (calculating). In addition, the management flag for the "image allocation pointer" is "not yet".
To the element that becomes. When the end of the processing area is reached, the processing is returned to the beginning and the search is performed, and the element which is first "not yet" is advanced. If there is no element that is "not yet", the image allocation pointer is set to "empty". Also, the "number of remaining processing areas" is updated ("number of remaining processing areas"-"size of next processing area") (step ST5). Next, in the processing area allocation operation, the processing restriction amount calculation means 7 is instructed to calculate the “processing restriction amount” for the selected “next processing area” (step ST6). The processing limit amount calculating means 7 obtains the maximum number of processings per unit area from the maximum processing amount setting means 3 per unit area, and from the “next processing area” instructed by the image area allocating means 2, in the target area. "Processing limit" is calculated. In the processing area allocation operation, the processing limit calculation means 7
After receiving the calculation result of the "processing restriction amount" from the above, the processing load calculation means 8 of the request source is notified of the information of the "next processing area" and the "processing restriction amount" (step ST7). With this notification, a series of processing area allocation operations are completed, and the state returns to the "waiting for request" state.

A case in which the "number of remaining processing areas" is 0 when the "number of remaining processing areas" is checked will be described. in this case,
First, the requesting processing load calculating means 8 is instructed to execute the "primary stop". Further, the in-execution flag of the processor corresponding to the request source processing load calculating means 8 is set to "OFF" (step ST8). Next, it is checked whether all the execution flags of the processor are "OFF" (step ST9). Here, the running flag is "ON" for even one
If there is any processor that is in the state, it returns to the "waiting for request" state. On the other hand, when all the running flags are “OFF”, the processing load calculating means 8 of all the processors are notified of “execution completed” (step ST10).

The operation in the case where the processing load calculating means 8 requests "reassignment" in the "request waiting" state will be described. In this case, first update the "number of remaining processing areas" (set to "number of remaining processing areas" + "size of reallocation area")
(Step ST11). Further, the management flag of the execution management table for each processing unit corresponding to the area for which "reallocation" is requested is set to "not yet" (from the total) (step ST1).
2). Here, if the image allocation pointer is "empty",
The image allocation pointer is set at the beginning of the area in which the management flag is updated to "not yet" (step ST13). Next, the in-execution flag of each processor is checked, and if there is a processor that is "OFF", "restart execution" is notified to the corresponding processing load calculation means 8 (step ST1).
4). Also, the in-execution flag corresponding to the processor that has notified the resumption of execution is set to "ON" (step ST1).
5). After this, the state returns to the "waiting for request" state.

The operation of the processing load calculation means 8 will be described with reference to FIG. The processing load calculating means 8 first receives an instruction to start execution from the image area allocating means 2 (step ST21), and requests the next processing (step ST2).
2) Wait for the answer (step ST23). here,
When the answer is the notification of the information of the "next processing area" and the "processing limit amount", a series of processing execution operations are started. In the process execution operation, first, within the processing area specified in "Next processing area",
The estimation calculation of the scattered point number is executed for each row (the estimated scattered point number is calculated) (step ST24). In the calculation of the estimated number of scattered points, the processing portion 12 that determines the processing load of the image processing program 11 is used. In this calculation, the maximum number of scattered points per line of the maximum processing amount setting means 3 per unit area is referred to so that the maximum value of the estimated scattered points in each line is equal to or less than this maximum scattered point. next,
The estimated number of scattering points in each row is added one by one in order from the first row of the "next processing area", and this sum is compared with the "processing limit amount" (steps ST25 to ST29). Here, a line whose sum exceeds the "processing limit amount" is defined as a "boundary line" (step ST30). The area from the beginning of the "next processing area" to the "boundary row" (including the "boundary row") is set as the "super-resolution execution area" (step ST31). After setting the "super-resolution execution region", the region from the line next to the "boundary line" to the last line of the "next processing region" is set to the "unprocessed region" (see FIG. 8) (step ST33). ). If the "boundary row" is the last row of the "next processing area" (the "reallocation area" is empty), the process proceeds to the next super-resolution processing execution (step ST32). On the other hand, when the “boundary line” is not the last line of the “next processing region” (the “reallocation region” exists), the image region allocation unit 2 is requested to reallocate the “unprocessed region” ( Step ST34). After notifying this request, the process proceeds to the next execution of super-resolution processing. In the execution of the super-resolution processing, the image processing execution means 9 is instructed to execute the super-resolution processing for the "super-resolution execution area" (step ST35). The image processing execution means 9 uses the processing portion 13 of the image processing program 11 whose load changes dynamically to execute the super-resolution processing. After that, after the completion of execution of the super-resolution processing in the image processing execution means 9, the processing execution operation is ended, the process returns to "request for next processing" and the processing is started (step ST36).

When the reply to the request for the next processing is "primary stop" of execution, the state is "waiting for instruction" (step S).
T37). When the "restart execution" is instructed from the image area allocating means 2 in the "waiting instruction" state, the process returns to the "request for next process" again to restart the process. On the other hand, in the "waiting for instruction" state, when the "execution completed" is instructed by the image area allocating means 2, the processing is ended (step ST38).

A case where the sample image shown in FIG. 9 is processed by a computer having a four-processor configuration in the apparatus of the first embodiment will be described. In FIG. 9, the processing is 1, 3, 5,
In the 7th line, the number of estimated scattering points is 100 and the processing is concentrated,
In all the other rows, the processing is small with an estimated number of scattered points being 1. Here, the calculation time for the estimated number of scattering points is 1, and the number of scattering points is 1.
The time required for super-resolution processing per piece is one.
Hereinafter, the unit of the processing time here is set to "T". When the process of FIG. 9 is sequentially executed, it takes 600T. Note that, in FIG. 9, the estimated scatter points and the processing load (processing time) of each line are clearly shown, but this is a value found after the estimation calculation result of the scatter points, and a value that cannot be known before the start of the calculation. Is. For comparison, first, a case will be described in which processing is simply assigned to each processor based on the size of the data area. Here, Guided-Sel
An example of f-scheduling (case executed without changing the allocation of the first processing area; see FIG. 7) will be described. In this case, the processor that first requested the allocation allocates the area from the first row to the 25th row. In this case, this processor intensively executes about 75% of the entire processing, and takes about 450T. On the other hand, the other processors are sequentially assigned the remaining areas. Here, since the load is constant, the other three processors are
Each of them will perform about 8% of the whole process,
It takes only 48T. As described above, when the area allocation in the general parallel processing is adopted and the initial allocation is not changed, the amount of processing allocated to each processor may not be maintained evenly during parallel calculation execution. At this time, the total execution time is the processing time of the processor that took the longest time.
T also takes time (only 1.3 times faster). If the processing amount cannot be kept uniform, the execution efficiency is reduced in this way.

On the other hand, in the apparatus of the first embodiment, the following processing is performed. As means for making various settings, the number of used processors of the used processor number setting means 4 is “4”, and the size of the image processing area of the processed image information setting means 5 is “100”.
Set to line. In the maximum throughput setting unit 3 per unit area, the maximum estimated number of scattering points per line is set to "10".
It is assumed that it is set to "0". Also, the mounting image is
It becomes like FIG. In the image area allocating means 2, the process proceeds according to the above-mentioned flowchart of FIG. First, initialization is performed based on the data set in each means. Set the running flags of processors 1 to 4 to "ON",
The "remaining processing area number" is set to 100, the management flags of lines 1 to 100 of the execution management table are set to "not", and the "image allocation pointer" is set to the first line. The area division method setting means 6
It is assumed that "(" remaining processing area number "+" used processor number "-1) /" used processor number "" is notified as the determination method of the size of the "next processing area". 1 to 8-4 are instructed to start execution, and the processing load calculation means 8-1 to 8-4 proceed with the processing in accordance with the flowchart of Fig. 3. Here, an instruction to start execution is given from the image area allocation means 2. In response, the "next processing request" is sequentially issued to the image area allocating means 2. Here, in order to simplify the explanation, first, the processing load calculation unit 8-1 first makes a “next processing request” to the image area allocation unit 2, and subsequently, at substantially the same time, the processing load calculation unit 8-2. , Processing load calculation means 8-
It is assumed that the "next processing request" is made in the order of 3, the processing load calculating means 8-4. Further, the processing load calculation means 8-1 to 8-1
8-4, at the time “0T”, performs the processing all at once “request for next processing”
I went. First, at time “0T”, the image area allocating means 2 sends the “next processing request” from the processing load calculating means 8-1.
To receive. Initially, the "remaining processing area number" is 100, and 25 rows from the beginning and 1 to 25 rows are selected as the "next processing area" according to the area dividing method of the area dividing method setting means 6. Next, the "remaining processing area number" is updated to 75, the management flags from the 1st row to the 25th row are updated to "total", and the "image allocation pointer" is updated to the 26th row. After that, the processing limit amount calculation means 7 is caused to calculate the “processing limit amount”. In this example,
"The size of the next processing area" * 4-1 "is the processing limit amount"
Will be calculated as Since the “size of the next processing area” is 25, the “processing restriction amount” is 99 here.
After calculating the “processing limit amount”, the processing load calculating means 8-1
"Next processing area" (1 to 25 lines) and "processing limit amount" 99
Is notified and the state returns to the "waiting for request" state. In the image area allocating means 2, the processing load calculating means 8 is further added at time “0T”.
-2, processing load calculating means 8-3, processing load calculating means 8-
4 receives the "next processing request". Here, the "next processing area" is allocated by the same procedure as when the processing load calculation means 8-1 makes a request. The processing load calculation unit 8-2 has the "next processing region" (lines 26 to 44) and the "processing limit amount" 75, and the processing load calculation unit 8-3 has the "next processing region" (lines 45 to 45). 58) and the "processing limit amount" 55, and the processing load calculation means 8-4 stores the "next processing area" (lines 59 to 6).
(Line 9) and "process limit amount" 43, and "wait for request"
Return to the state. As a result of the processing here, the "remaining processing area number" is updated to 31, the management flag from the 26th line to the 69th line is updated to "total", and the "image allocation pointer" is updated to the 70th line.

In the processing load calculation means 8-1, the time "0"
At “T”, the “next processing area” (lines 1 to 25) and the “processing limit amount” 99 are received from the image area allocating means 2 as a reply to the “next processing request”. First, the estimated number of scattering points in each line of the "next processing region" is calculated. Here, it takes "25T" to calculate the estimated number of scattered points for 25 lines. Next, the sum of the estimated number of scattered points in each row is calculated and compared with the “processing limit amount” 99. Here, the estimated number of scattering points is 100 in the first line, and the first line is the “boundary line”. The "super-resolution execution area" is only the first row. Naturally, since the "boundary line" is not the last line, lines 2 to 25 are set as "unprocessed regions" and the image region allocating means 2 is requested to reallocate. Here, a request for reallocation to the image area allocating means 2 is made at time “25T” when the calculation of the estimated scattered points is completed. After the reallocation request, the processing load calculation means 8-1 sends the image processing execution means 9-1 to the image processing execution means 9-1.
Instruct execution of the super-resolution processing of the "super-resolution execution area" (first line). The processing load calculation means 8-1 waits for the end of this processing until time "125T". In the processing load calculation means 8-2, the "next processing area" (26
(Line to line 44) and the "processing limit amount" 75 are received. The processing proceeds in the same procedure as the processing load calculating means 8-1, but here, the sum of the estimated scattering points of each row increases to 19 even if the calculation is performed up to the last row. Therefore, the reallocation request is not issued, and all of the 26th to 44th rows which are the "next processing area" are set to the "super-resolution execution area". After setting the "super-resolution execution area", the 26th line to the image processing execution means 9-2.
Instruct execution of super-resolution processing of 44 lines. Here, the end of this processing is waited until time "38T". The processing load calculation means 8-3 receives the "next processing area" (lines 45 to 58) and the "processing limit amount" 55 at time "0T". Here, as in the processing load calculation means 8-2, the cumulative total of the number of estimated scattering points increases to only 14 and a value smaller than the processing limit amount. Therefore, lines 45 to 58 are set as the “super-resolution execution region”. Then, the image processing execution means 9-3 is instructed to execute the super-resolution processing. The end of this processing is waited until time "28T". The processing load calculation means 8-4 receives the “next processing area” (lines 59 to 69) and the “processing limit amount” 43 at time “0T”. In this case as well, since the cumulative total of the estimated number of scattered points increases only to a value smaller than the processing limit amount of 11, the 59th to 69th lines are set in the "super-resolution execution region" and the image processing execution means 9-4 is operated. Instruct execution of resolution processing. The end of this process will be waited until time “22T”.

Since the execution of the super-resolution processing is completed at the time "22T", the processing load calculating means 8-4 gives the "next processing request".
I do. Here, in the image area allocating means 2, the time "0T"
In the same procedure as in the above case, the processing load calculation means 8-4 is notified of the "next processing area" (lines 70 to 77) and the "processing limit amount" 31, and the state returns to the "request waiting" state. The "remaining processing area number" is updated to 23, the management flag from the 70th line to the 77th line is updated to "total", and the "image allocation pointer" is updated to the 78th line. Further, the processing load calculation means 8-4 which receives this also
Proceed with the same procedure as last time. In this case as well, since the cumulative total of the estimated number of scattered points increases only to a value smaller than the processing limit amount of 8, the 70th to 77th lines are set in the "super-resolution execution region", and the image processing execution means 9-4 is operated. Instruct execution of resolution processing. The processing load calculation means 8-4 waits for the end of this processing until time "38T".

At time "25T", the processing load calculation means 8-
1 requests re-allocation of 2 to 25 lines (24 lines).
Here, the image area allocating means 2 sets the “number of remaining processing areas” to 23.
To 47 (24 lines are added), and the management flags of lines 2 to 25 are set to "not yet". Here, processor 1
Since all of the in-execution flags of 4 are "ON", the process related to resumption of execution is not performed and the state returns to the "waiting for request" state.

Since the execution of the super-resolution processing is completed at time "28T", the processing load calculation means 8-3 gives "the next processing request".
I do. Here, the image area allocating means 2 basically advances the processing in the same procedure as at the time “0T”. However, since the reallocation request is received, the "remaining processing area number" is updated from 23 at the time "25T" to 47 at the time "25T". Therefore, the "size of next processing area" is 12
(Note that in the “number of remaining processing areas” 23, the size of “the size of the next processing area” was 4.) As a result, the image area allocating means 2 notifies the processing load calculating means 8-3 of the "next processing area" (lines 78 to 89) and the "processing limit amount" 47, and returns to the "request waiting" state. The "remaining processing area number" is updated to 35, the management flags from the 78th line to the 89th line are updated to "total", and the "image allocation pointer" is updated to the 90th line. On the side of the processing load calculation means 8-3 which receives this, the processing proceeds in the same procedure as the previous time. In this case as well, the cumulative total of the estimated number of scattered points increases to 12 which is smaller than the processing limit amount. Therefore, lines 78 to 89 are set in the "super-resolution execution region" and the image processing execution means 9-3 is operated. Instruct execution of resolution processing. The processing load calculation means 8-3 waits for the end of this processing until time "52T".

Since the execution of the super-resolution processing is completed at the time "38T", the processing load calculation means 8-2 gives the "next processing request".
I do. In the image area allocating means 2, the "next processing area" (9
Lines 0 to 98) and the "processing limit amount" 35 are notified to the processing load calculation means 8-2, and the state is returned to the "request waiting" state. The "remaining processing area number" is updated to 26, the management flag from the 90th line to the 98th line is updated to "total", and the "image allocation pointer" is updated to the 99th line. On the side of the processing load calculation means 8-2 that receives this, the processing proceeds in the same procedure as the previous time. Since the cumulative total of the estimated number of scattered points increases only to a value smaller than the processing limit amount of 8, the 90th to 98th lines are set to the "super-resolution execution region",
The image processing execution means 9-2 is instructed to execute the super-resolution processing. The processing load calculation means 8-2 waits until the time "56T" until the end of this processing.

The processing load calculation means 8-4 also receives the time "38".
Since the execution of the super-resolution processing ends at "T", a "next processing request" is issued. Here, it is assumed that it is received immediately after the completion of the processing for the request of the processing load calculation means 8-2. Here, the image area allocating means 2 basically advances the processing in the same manner as when the request from the processing load calculating means 8-2 was made. As a result, the "size of the next processing area" is 8. at this point,
Since the "image allocation pointer" is the 99th line, 99,100
After selecting a row, rows 2 to 7 are selected as the "next processing area". Therefore, the "next processing area" (lines 99 to 100 and 2 to
7)) and the "processing limit amount" 31 to the processing load calculation means 8-4.
To the "request waiting" state. The "remaining processing area number" is updated to 18, the management flags of 99 to 100 lines and 2 to 7 lines are set to "total", and the "image allocation pointer" is updated to the 8th line by making one round. Receiving this, the processing load calculation means 8-4
On the side, the process proceeds in the same procedure as the previous time. Here, 9
When the cumulative total of the estimated scattered points is calculated in the order of 9,100,2,3, the estimated scattered points becomes 103 in the cumulative total of up to three lines, which exceeds the processing limit amount. Here, the third line is the “boundary line”. Therefore, the processing load calculation means 8-1 described above is used.
Similarly, the 4th to 7th lines are set to the "unprocessed area" and the image area allocating means 2 is requested to reallocate. Here, a reassignment request to the image area allocating means 2 is made at time “46T” when the calculation of the estimated scattered points is completed. Also, 99,10
Lines 0, 2, and 3 are set in the "super-resolution execution area", and the image processing execution means 9-4 is instructed to execute the super-resolution processing. The processing load calculation means 8-4 indicates the end of this processing at time “149”.
You will have to wait until "T".

At time "46T", the processing load calculating means 8-
4 requests re-allocation of lines 4 to 7 (for 4 lines). Here, the image area allocating means 2 performs the same processing as at the time “25T”, updates the “remaining processing area number” from 18 to 22 (adds 4 rows), and sets the management flags of 4 rows to 7 rows. After setting it to "not yet", it returns to the "request waiting" state.

At time "52T", the super-resolution processing ends, and the processing load calculation means 8-3 makes a "next processing request". Here, both the image area allocating means 2 and the processing load calculating means 8-3 proceed as in the previous time “52T”. In the image area allocating means 2, the "remaining processing area number" is 16 and 8 to
The management flag of line 13 is "total" and "image allocation pointer"
Is updated on the 14th line. "Next processing area" (8 to 13
Line) and the “processing limit amount” 23, the processing load calculation means 8
In -3, since the cumulative total of the estimated number of scattered points increases to 6, lines 8 to 13 are set as the "super-resolution execution area" and the image processing execution means 9-3 is instructed to execute the processing. The end of this process is waited until time "64T".

Thereafter, in the same procedure at time "52T",
At time “56T”, the processing load calculation means 8-2 is 14-17.
At the time "64T", the processing load calculation unit 8-2 makes 18
~ 20 rows, the processing load calculating means 8-3 is assigned 21 to 22 rows at the time "64T", and the processing load calculating means 8-2 is assigned 23 to 24 rows at the time "70T". It will be. As a result, in the image area allocating means 2, the "remaining processing area number" is updated to 4, the management flags of lines 14 to 24 are updated to "total", and the "image allocation pointer" is updated once again to the 4th row. It Further, the processing load calculation means 8-2 waits until the time "74T", and the processing load calculation means 8-3 waits until the time "70T" to complete the super-resolution processing.

At time "70T", the processing load calculation means 8
-3 is executed by assigning 4 lines (only 1 line). As a result, in the image area allocating means 2, the "remaining processing area number" is updated to 3, the management flag of the 4th row is updated to "total", and the "image allocation pointer" is updated to the 5th row. In addition, the processing load calculation unit 8-3 waits for the completion of execution of the super-resolution processing until the time “72T”.

At time "72T", the processing load calculation means 8
-3 will be executed by assigning 5 lines. As a result, in the image area allocating means 2, the “number of remaining processing areas” is 2
In addition, the management flags of the 5th row are updated to “total”, and the “image allocation pointer” is updated to the 6th row. Further, the processing load calculation means 8
-3 waits for the completion of execution of the super-resolution processing until time “173T”.

Thereafter, the processing load calculating means 8-2 allocates and executes 6 lines at time "74T" and 7 lines at time "76T". As a result, in the image area allocating means 2, the "remaining processing area number" is 0, and the management flags of the 6th and 7th rows are "total". Since all the management flags are “total”, there is no line to be pointed to by the “image allocation pointer” (the next time when a reallocation request is made, it is set again).
In addition, the processing load calculation unit 8-2 waits for the completion of execution of the super-resolution processing until time “177T”.

At time "125T", the processing load calculation means 8
-1 makes a "next processing request". In the image area allocating means 2, since the “remaining processing area number” is 0, the processing load calculating means 8
-1 is instructed to pause, and the in-execution flag of the processor 1 is set to "OFF". At this point, processors 2-4
Since the in-execution flag of is ON, the process returns to the request waiting state. In response to the "next processing request", the processing load calculation means 8-1 instructed to pause will wait in the "instruction waiting state".

At time "149T", the processing load calculation means 8
-4, at time “173T”, processing load calculation means 8-3
Makes the "next processing request". The image area allocating means 2 performs the same processing as that for the processing load calculating means 8-1 at time “125T”, and instructs the processing load calculating means 8-4 and the processing load calculating means 8-3 to temporarily stop. , The in-execution flags of the processor 4 and the processor 3 are turned off. At this point, all the execution flags are not yet set to "OFF", and the process returns to the "request waiting state". Further, in response to the "next processing request", the processing load calculation means 8-4 and the processing load calculation means 8-3, which have been instructed to pause, are in the "instruction waiting state".
You will have to wait.

At time "177T", the processing load calculation means 8
-2 makes a "next processing request". Here, the image area allocating means 2 instructs the processing load calculating means 8-2 to temporarily stop, and sets the execution flag of the processor 2 to “OFF”. Here, since all the execution flags are “OFF”, “execution completed” is notified to all the processors, and the image area allocation unit 2 ends the processing. Processing load calculation means 8-2
Then, since the temporary stop is instructed in response to the "next processing request", the process waits in the "instruction waiting state". after,
Upon receiving the notification of "execution completed" from the image area allocating means 2, the processing load calculating means 8-2 ends the processing. The processing load calculation unit 8-1, the processing load calculation unit 8-3, and the processing load calculation unit 8-4, which have been waiting in the “instruction waiting state”, also receive the notification of “execution completed”, and terminate the respective processes.

As described above, in the example of the image of FIG. 9 this time, in the apparatus of the first embodiment, the processing can be completed in only 177T (the speed can be increased by about 3.4 times).

As described above, according to the first embodiment, the processing load assigned to each processor is monitored on the basis of the number of scattering points serving as the load reference, and the processing by a certain processor exceeds the limit amount. In this case, by reallocating the processing area allocated to the processor that exceeds the limit amount, it is possible to obtain the parallel image processing apparatus capable of efficiently executing the algorithm for image processing in which the processing load changes depending on the input image. . Also, by adjusting the size of the area to be allocated according to the request for reallocation of the processing area, the allocation of the processing area to each processor can be made more uniform, and the algorithm for image processing can be executed efficiently. There is an effect that an image processing device capable of being obtained can be obtained. This allows Gu
Even in the case where sufficient load distribution cannot be performed by simple dynamic area division such as ided-Self scheduling,
There is an advantage that the algorithm for image processing can be executed more efficiently. In each processor, the calculation to judge the processing load in each row is performed first, and if the limit is exceeded,
After requesting the allocation of the image area again, the main part (here, super-resolution processing based on the estimated number of scattered points) in the area that the user is in charge of is executed. As a result, there is an advantage that reallocation to another processor can be performed efficiently.

In the above-described first embodiment, the example of the super-resolution processing of the SAR image using MUSIC has been described. However, other than this image processing, as long as the image processing satisfies the following condition, The same effect as that of the first embodiment can be obtained. An algorithm for image processing in which the processing load changes depending on the input image, and is calculated based on the result of the first half load calculation unit and the part that calculates the element for determining the processing load for each unit area (first half load calculation unit) It can be divided into the part that performs (second half process execution part). Further, in the calculation of the unit area, the first half load calculation unit can calculate an element for determining the processing load within a certain processing time, and the second half process execution unit can complete the processing in a time proportional to the calculated element.

In the first embodiment, the parallel image processing apparatus 1 is mounted independently as the parallelization support library as the mounting method, but it is a platform O.
Even if it is implemented by a method of incorporating into S or a parallelized library, a method of incorporating into an image processing program, etc.
The same effect as can be obtained. Platform 1
Although the shared memory type SMP has been described as an example of the H / W configuration of 0, the first embodiment does not depend on a specific computer configuration, and in an arbitrary parallel computer (H / W configuration), The same effect as that of the first embodiment can be obtained. Further, even in the shared memory type SMP, it can be realized by dividing the means realized as a common part among the processors into the respective processors and operating the same, as in the first embodiment. It is possible to exert an effect.

In the first embodiment, an example in which "(" remaining processing area number "+" used processor number "-1) /" used processor number "" is used as a method of calculating the size of the "next processing area" However, any calculation method other than this may be adopted as long as the calculation method can cope with a reallocation request and an increase in the number of image areas to be processed in the middle. Therefore, it is possible to achieve the same effect as in Embodiment 1. Further, as the method of selecting the “next processing area”, the method of sequentially selecting from the beginning of the image area has been described, but the unprocessed area remains. Any area selection method can be adopted as long as it is a method that can be allocated without any effect, and the same effect as that of the first embodiment can be obtained. This is the same for the management flags in all processing units in the execution management table, and any management method can be adopted as long as it is a method that can be allocated without leaving an unprocessed area. The same effect as that of the first embodiment can be obtained.

In the above-described first embodiment, the end determination method and
It controlled the re-execution of the processor whose processing was once completed. These are only examples for explanation, and any method can be adopted as long as it is determined that all the processing is completed and the processor that has once completed the processing is allocated a processing area again and can be executed. Therefore, the same effect as that of the first embodiment can be obtained. For example, in the image area allocating means 2, when the “next processing area” is received from the processing load calculating means 8, the management flag of the execution management table corresponding to the “next processing area” passed immediately to the processor is set to “complete”. To In addition, when the "reallocation request" is received, the range of the "next processing area" that was immediately passed to the processor that has not been reallocated is set to "completed". Furthermore, when the remaining processing area is 0 and the next processing request is received,
If all the management flags are “complete”, the end is passed, and if even one management flag is not “complete”, an empty (size 0) “next processing area” is passed to the processing load calculating means 8. When the processing load calculation means 8 receives an empty "next processing area", it waits for a certain period of time and outputs the "next processing area" again. When the termination is instructed, the processing in each processor is terminated.
Even with such a mounting method, the same effect as that of the first embodiment can be obtained.

In the above-described first embodiment, the processing load calculation means 8 first calculates the estimated scattering points of all the assigned "next processing areas" and then calculates the "boundary line". This is a method of determining the "boundary line" by calculating the estimated scattered points for each row and calculating the cumulative estimated scattered points up to that row, and comparing the obtained cumulative estimated scattered points with the processing limit amount. Even if it is realized by, it is possible to obtain the same effect as that of the first embodiment. In this case, it is not necessary to calculate the estimated scattered points in the range where the cumulative estimated scattered points exceed the processing limit amount and the reallocation request is made. In addition, although the determination of the “boundary line” is described as an example in which the line exceeds the processing limit amount, the same effect as that of the first embodiment can be obtained even when the line before the processing limit amount is exceeded. You can In this case, a certain range (for example, only the first line) from the beginning of the allocated "next processing area" is subject to restrictions such that it is always executed, or the value set for the processing limit amount is adjusted. Good to do.

In the first embodiment, the image of FIG. 9 and the image of FIG.
Although the computer configuration of 0 has been described, these are 1 for explanation.
This is an example, and the same effect as that of the first embodiment can be obtained with any image data and any configuration (any number of processors, etc.).

Embodiment 2. FIG. 11 is a block diagram showing a parallel image processing apparatus according to the second embodiment of the present invention. In the figure, 14 is a load calculation type area division method setting means which extends the function of the area division method setting means 6 in FIG. is there.
Other configurations are the same as those in FIG.

Next, the operation will be described. In the second embodiment, the load calculation type area division method setting unit 14 refers to the maximum processing load amount per unit area of the maximum processing amount setting unit 3 instead of the area division method setting unit 6. The method of dividing the image area is different from that of the first embodiment. Same as the first embodiment except that the “next processing area” is determined based on the division method of the image area designated by the maximum processing amount setting unit 3 per unit area (operation procedure of other means). Is.

In the first embodiment, the Guided-Sel is used.
f Refer to the scheduling, “(“ Number of remaining processing areas ”+
"Number of used processors" -1) / "Number of used processors""
Has selected the method for determining the size of the "next processing area". Here, how much the maximum processing load in the unit area is is not considered. In the second embodiment, the load calculation type area division method setting unit 14 is referred to with reference to the maximum processing load amount specified by the maximum processing amount setting unit 3 per unit area.
Determines the method of calculating the "next processing area" size. In the load calculation type area division method setting means 14, per line,
An arbitrary "next processing area" size calculation method can be designated based on the maximum calculable estimated scattering points. here,
The maximum processing amount can be used for the selection condition of the calculation formula and the calculation formula itself. For example, consider a case where the maximum number of estimated scattering points per line that can be calculated is about several (for example, a few) as the maximum amount of processing in a unit area. In this case, it can be inferred that the difference in processing load due to the difference in the calculated estimated scattering points in each row is relatively small. At this time, for example, referring to simple scheduling,
"(" Number of remaining processing areas "+" Number of used processors "-1) /
Select the method to determine the size of the "next processing area" in "Number of processors used". On the other hand, let us consider, for example, a case where the maximum number of estimated scattering points per line that can be calculated is several hundreds (for example, a few hundreds) as the maximum processing load amount in a unit area. In this case, it can be inferred that the difference in processing load due to the difference in the calculated estimated scattering points in each row is relatively large. At this time, it is desired to detect as early as possible whether a fatal processing load has occurred. Also, in order to balance the processing load due to reallocation, it is important to allocate rows of a certain size or more to each processor.
Therefore, the size of the "next processing area" is determined by the following method, for example. First, the “temporary processing number” is determined by “(“ remaining processing area number ”+“ used processor number ”−1) /“ used processor number ””. When the "temporary processing number" is larger than 5 lines, the size of the "next processing area" is set to "the value of the" temporary processing number "" where the "temporary processing number" is 5 lines or less and the "remaining processing area number" Is larger than 10 lines, the size of the “next processing area” is set to “5 lines”. When the “temporary processing number” is 5 lines or less and the “remaining processing region number” is more than 5 lines and 10 lines or less, the size of the “next processing region” is set to “the number of remaining processing regions” / 2. When the "number of processes" is 5 lines or less and the "number of remaining processing regions" is 5 lines or less, the size of the "next processing region" is set to "the value of the" number of remaining processing regions "".

As described above, according to the second embodiment, the method of calculating the size of the processing area to be assigned to each processor can be determined based on the maximum processing load amount in the unit area. Here, there is an advantage that the selection condition of the calculation formula can be selected on the basis of the maximum processing load amount in the unit area when the size assigned to each processor is determined. Further, the calculation formula itself has an advantage that the maximum processing load amount in the unit area can be used. As a result, there is an effect that it is possible to obtain a parallel image processing device capable of allocating a processing region of an appropriate size to each processor according to the maximum processing load amount in the unit region.

In the second embodiment described above, the calculation method using the "temporary processing number" and the like is described based on the simple scheduling. However, even with the calculation method other than this, the maximum processing in the unit area is performed. An arbitrary calculation method considering the load amount can be adopted, and the same effect as that of the second embodiment can be obtained. Although the "provisional number of processes" is 5 lines or more as the criterion for selecting the calculation method, any criterion other than this may be used as an arbitrary criterion considering the maximum processing load amount in the unit area. This can be adopted, and the same effect as that of the second embodiment can be obtained. Further, the values of the 5th row, the 10th row, and the like can be set to arbitrary values, and the same effect as that of the second embodiment can be obtained.

Embodiment 3. FIG. 12 is a block diagram showing a parallel image processing apparatus according to Embodiment 3 of the present invention. In the figure, 15 is a processing load measuring means, and 16 is an actual measurement correspondence in which the function of the area division method setting means 6 in FIG. 1 is expanded. The pattern area division method setting means 17 is an actually-measured processing restriction amount calculation means in which the function of the processing restriction amount calculation means 7 in FIG. Other configurations are the same as those in FIG.

Next, the operation will be described. In the third embodiment, the processing load measuring means 15 measures the execution time when the processing load required for the calculation per unit area is the maximum and the execution time when the processing load is the minimum, and the area division method setting means 6 is used. Instead of, the actual measurement correspondence area division method setting means 16 refers to the actual measurement time of the processing load actual measurement means 15 to determine the image area division method, and instead of the processing limit amount calculation means 7, actual measurement correspondence It differs from the first embodiment in that the mold processing restriction amount calculating means 17 calculates the processing restriction amount with reference to the measured time of the processing load measuring means 15.

In the third embodiment, the processing load measuring means 15 uses the maximum processing load amount (in this example, the maximum value of the estimated scattering points) designated by the maximum processing amount setting means 3 per unit area as a reference. , The maximum execution time (execution time when the estimated scatter score is maximum) and the minimum execution time (execution time when the estimated scatter score is 0) required for calculation per unit area (1 line) are measured. Is recorded.

In the first embodiment, the Guided-Sel is used.
f Refer to the scheduling, “(“ Number of remaining processing areas ”+
"Number of used processors" -1) / "Number of used processors""
Has selected the method for determining the size of the "next processing area". Further, in the second embodiment, the load calculation type area division method setting means 14 calculates the “next processing area” size with reference to the maximum processing load amount specified by the maximum processing amount setting means 3 per unit area. I had decided on a method. In the third embodiment, the actual measurement correspondence area division method setting unit 16 calculates the “next processing area” size with reference to the execution time recorded by the processing load actual measurement unit 15. The measurement-corresponding area division method setting means 16 can specify an arbitrary “next processing area” size calculation method based on the maximum and minimum execution times that change per unit area (one row). Here, the selection condition of the calculation formula and the calculation formula can be determined based on the maximum and minimum execution times.

In the third embodiment, the actually-measured processing limitation amount calculation means 17 calculates the processing limitation amount with reference to the execution time recorded by the processing load measurement means 15. In the processing limit amount calculation means 7 of the first and second embodiments,
The processing limit amount was calculated by "the size of the next processing area" * 4-1. Here, the size of the given processing area and the maximum value specified by the maximum processing amount setting means 3 per unit area are calculated. It has been assumed that an appropriate expression is specified from the processing amount.On the other hand, in the third embodiment, the maximum per line is set with reference to the execution time recorded by the processing load measuring unit 15. It is possible to specify an arbitrary formula for calculating the processing limit amount based on the minimum execution time.

The operation procedure of the other means is the same as that of the first or second embodiment.

As described above, according to the third embodiment, the maximum and minimum calculation times in the unit area are measured, and the size of the processing area assigned to each processor is calculated based on the measured value. The method can be determined. Here, when deciding the size to allocate to each processor,
There is an advantage that the selection condition of the calculation formula and the calculation formula itself can be determined based on the maximum and minimum calculation times in the unit area. As a result, there is an effect that it is possible to obtain a parallel image processing apparatus capable of allocating a processing region of an appropriate size to each processor based on the maximum and minimum calculation times in the actually measured unit region. Further, in the present device, the calculation formula for obtaining the “processing restriction amount” can be determined based on the actual measurement result of the maximum and minimum calculation times in the unit area. Thus, it is possible to obtain a parallel image processing device capable of calculating the “processing limit amount” corresponding to the “next processing region” assigned to each processor on the basis of the maximum and minimum calculation times in the unit region. effective.

Fourth Embodiment 13 is a block diagram showing a parallel image processing apparatus according to Embodiment 4 of the present invention. In the figure, 18 is a processing block setting means, and 19 is a processing block corresponding to an extended function of the processing load calculating means 8 in FIG. A type processing load calculating means, 20 is a processing block corresponding type image processing executing means which is obtained by expanding the function of the image processing executing means 9 in FIG. Other configurations are the same as those in FIG.

Next, the operation will be described. In the fourth embodiment, the processing block corresponding processing load calculating means 19 and the processing block corresponding image processing executing means 20 preferentially perform the processing of the area designated by the processing block setting means 18. Different from 1. The operation procedure of other means is the same as that of the first embodiment.

Processing block corresponding processing load calculating means 19
In, an image processing area of arbitrary size and shape can be designated as a “block”. In addition, priority calculation can be specified in units of this “block”. The processing block corresponding processing load calculating means 19 and the processing block corresponding image processing executing means 20 are processing block setting means 1
Within the range of the “block” designated by 8, the calculation designated by the processing block setting means 18 is preferentially executed.

First, an example in which a "block" is larger than a unit area (row) will be described. For example, two lines from the beginning of the "next processing area" as a "block", and as a priority calculation, "the part that performs the process of estimating the scatter point based on the eigenvalue, and the resolution improvement of the SAR image based on the estimated scatter point" The following shows the case in which "continuous calculation with the part for performing processing" is specified. Here, if the size of the “next processing area” is two lines or more, and if the processing amount up to this point does not exceed the “limit processing amount” notified from the image area allocation unit 2, the processing block The corresponding processing load calculating means 19 first instructs the processing block corresponding image processing executing means 20 to execute calculation for the first two lines. After that, the processing load corresponding to the processing block 19 is "boundary line".
Detection, the reallocation request (when the “limit processing amount” is exceeded) and the “super-resolution execution area” are calculated for the third and subsequent rows from the beginning, and the “super-resolution execution area” is calculated. The execution is instructed to the processing block corresponding image processing executing means 20. In addition, for example, as a “block”, a “super-resolution execution area”
In this case, the case where "the part for performing the resolution improvement processing of the SAR image on the basis of the estimated number of scattering points" is designated as the priority calculation for the area for four rows of. In this case, the processing load corresponding to the processing block 19 calculates the “super-resolution execution area”, and then calculates the “super-resolution execution area”. 4 lines are selected every four lines, and a calculation execution instruction is given to the processing block corresponding image processing execution means 20. For example, when there are 12 lines of "super-resolution execution regions", the processing block corresponding image processing execution means 20 is instructed to select every four lines and execute the calculation in three times.

Next, an example in which the “block” is smaller than the unit area (row) will be described. For example, the range that can be stored in the cache line size as "block" (here,
"Assuming 8 pixels"), the case where "a portion for performing the resolution improvement processing of the SAR image based on the estimated number of scattering points" is designated as the priority calculation is shown. In this case, the image processing execution unit 20 corresponding to the processing block performs the calculation of “a portion for executing the resolution improvement processing of the SAR image based on the estimated number of scattering points” for each row designated in the “super-resolution execution area”. When performing, the calculation is repeated for each “8 pixels”.

As described above, according to the fourth embodiment, an image processing area having an arbitrary size and shape is “blocked”.
, And the calculation to be prioritized can be specified in “block” units. By appropriately designating this “block” and the calculation to be prioritized, it is possible to obtain a parallel image processing apparatus with high execution efficiency.

In the above-mentioned fourth embodiment, the “block” and the prioritized calculation are explained with an example in which two lines from the beginning and “a part for carrying out the resolution improving process of the SAR image based on the estimated number of scattering points” are designated. However, arbitrary priority calculation can be designated for a “block” having an arbitrary size and shape, and the same effect as that of the fourth embodiment can be obtained. In the above-described fourth embodiment, the description has been given as an extension to the first embodiment. However, the processing block setting means 18 and the processing block corresponding processing load calculating means in the apparatus of the second or third embodiment are added. 19. Even if the parallel image processing apparatus is configured by adding the image processing execution unit 20 corresponding to the processing block 19, the same effect as that of the fourth embodiment can be obtained.

Embodiment 5. FIG. 14 is a block diagram showing a parallel image processing apparatus according to Embodiment 5 of the present invention. In the figure, 21 is a cache information setting means and 22 is a processing block calculating means. Other configurations are the same as those in FIG.

Next, the operation will be described. In the fifth embodiment, the processing block calculation means 22 uses the cache information of the cache information setting means 21 to determine the size and shape of the “block” and the priority calculation in “block” units. Different from the form 4. Others are the same as those in the fourth embodiment.

For example, in the fourth embodiment, "block"
The area for four lines of the “super-resolution execution area” and the portion for performing the resolution improvement processing of the SAR image on the basis of the estimated number of scattering points are designated as the priority calculation. In the fifth embodiment, the processing block calculation means 22 stores the cache information of the cache information setting means 21 (size of cache line, control method, total capacity of cache, etc.).
An area (the number of rows) that is advantageous in cache access is calculated from the image size and pixel size information set by the processed image information setting unit 5 and defined as a “block”. The range in which the cache access is advantageous depends on the combination of the cache configuration and the image size (particularly, the line size) to be processed. The processing block calculation unit 22 can calculate the “block” for which the cache access is advantageous for each image size of the processing target (since the cache configuration is fixed on the same computer, the image size is changed here. Explanation).

Further, in the fourth embodiment, the range which can be accommodated in the cache line size as “block” (here, “8 pixels” is assumed), and the priority calculation is as follows:
The "portion where the resolution improvement processing of the SAR image is performed based on the estimated number of scattering points" was designated. In the fifth embodiment, the range within the line size of the cache is set to the cache information of the cache information setting means 21,
It is possible to automatically calculate and set the pixel size information set by the processed image information setting means 5.

As described above, according to the fifth embodiment, the size and range of the "block" can be set from the cache information and the processing target image information. This allows
There is an advantage that the specified range of "block" can be set more appropriately.

Sixth Embodiment FIG. 15 is a block diagram showing a parallel image processing apparatus according to Embodiment 6 of the present invention, in which 23 is a dynamic processing block calculating means. Other configurations are the same as those in FIG.

Next, the operation will be described. In the sixth embodiment, the dynamic processing block calculating unit 23 determines the size of the “block” in the processing block setting unit 18 from the size information of the processing target area designated by the processing block corresponding processing load calculating unit 19. The fourth embodiment differs from the fourth embodiment in that the calculation that gives priority to the shape and the “block” unit is dynamically changed and determined. Others are the same as those in the fourth embodiment.

For example, in the fourth embodiment, "block"
The area for four lines of the “super-resolution execution area” and the portion for performing the resolution improvement processing of the SAR image on the basis of the estimated number of scattering points are designated as the priority calculation. At this time, the “block” remains four lines regardless of the size of the area designated by the processing block corresponding processing load calculation means 19. On the other hand, in the sixth embodiment, the size of the “block” can be changed according to the size designated by the processing block corresponding processing load calculating means 19. For example, when the “super-resolution execution region” is a multiple of 4, the “block” size remains 4 lines, and when it is divided by 4 and 1 is left, the possible range is 4 lines and the last “block” size is set. When dividing into 1 line by 4 and leaving 2 remainders, the possible range is 4 lines and the last "block" size is 2.
If a line is divided by 4 and there is a remainder of 3, then the possible range is 4 lines, and finally, the "block" size can be set to 2 lines and 1 line. Further, the size of the “block” can be changed from the information of the remaining processing area held by the image area allocating means 2.

As described above, according to the sixth embodiment, the size and range of the “block” can be dynamically set based on the size of the assigned “next processing area”.
Further, the size and range of the “block” can be dynamically set also from the information of the remaining processing area. This allows you to "block"
There is an advantage that the specified range of can be set more appropriately.

Seventh Embodiment 16 is a block diagram showing a parallel image processing apparatus according to a seventh embodiment of the present invention. In the figure, 24 is the area division method setting means 6 in FIG.
It is a processing block corresponding type area division method setting means that extends the function of. Other configurations are the same as those in FIG.

Next, the operation will be described. In the seventh embodiment, processing block corresponding area division method setting means 24 is provided.
However, unlike the fourth embodiment, the area allocation method is determined in consideration of the size of the “block” designated by the processing block setting unit 18 and the priority processing. Others are the same as those in the fourth embodiment.

For example, in the fourth embodiment, "block"
2 lines from the beginning of the “next processing region”, as a priority calculation, “a part that performs the process of estimating the number of scattering points based on the eigenvalue, and a part that performs the resolution improvement process of the SAR image based on the estimated number of scattering points. "Continuous calculation with" was specified. However, in the fourth embodiment, the size of the processing area assigned to the processing block corresponding processing load calculating means 19 is determined without considering this “block”. On the other hand, in the seventh embodiment, the processing block corresponding area division method setting means 24 considers the size of the “block” designated by the processing block setting means 18 and the priority processing, and allocates the area. Can be determined. Therefore, for example, processing block corresponding processing load calculating means 19
It becomes possible to select an area allocation method in which the processing area to be allocated to is selected as an area larger than two consecutive rows as much as possible.

As described above, according to the seventh embodiment, it is possible to specify the method of determining the processing area to be assigned to each processor based on the size and range of the specified "block" and the priority calculation. And this gives
There is an effect that it is possible to obtain a parallel image processing device that can more appropriately execute the calculation of the range specified by the “block”.

Eighth Embodiment FIG. 17 is a block diagram showing a parallel image processing apparatus according to Embodiment 8 of the present invention. In FIG.
It is a processing block corresponding processing limit amount calculating means that extends the function of. Other configurations are the same as those in FIG.

Next, the operation will be described. In the eighth embodiment, processing block corresponding processing limit amount calculating means 25
However, in consideration of the size of the “block” designated by the processing block setting unit 18 and the process to be prioritized, the maximum processing load amount for the allocation area can be calculated according to the fourth embodiment.
Different from Others are the same as those in the fourth embodiment.

For example, in the fourth embodiment, "block"
2 lines from the beginning of the “next processing region”, as a priority calculation, “a part that performs the process of estimating the number of scattering points based on the eigenvalue, and a part that performs the resolution improvement process of the SAR image based on the estimated number of scattering points. "Continuous calculation with" was specified. However, in the fourth embodiment, the processing limit amount notified to the processing block corresponding processing load calculation means 19 is determined without considering this “block”. On the other hand, in the eighth embodiment, the processing block corresponding processing limit amount calculating means 25 takes into consideration the size of the “block” designated by the processing block setting means 18 and the priority processing, and the maximum processing for the allocation area. The load can be calculated. For this reason, when determining the processing limit amount for a certain "next processing area", the processing limit amount for the first two lines and the processing limit amount for the third and subsequent lines are set to different values, and continuous execution in the first two lines is performed. It is possible to do something that can be prioritized.

As described above, according to the eighth embodiment, the processing limit amount for the processing area assigned to each processor is arbitrarily specified based on the size and range of the specified "block" and the priority calculation. This makes it possible to obtain a parallel image processing device capable of more appropriately executing the calculation of the range designated by the “block”.

Ninth Embodiment FIG. 18 is a block diagram showing a parallel image processing apparatus according to Embodiment 9 of the present invention. In the figure, the cache information setting means 21, processing block calculating means 22, and dynamic processing are added to the configuration shown in FIG. Block calculation means 23, processing block correspondence area division method setting means 24, processing block correspondence processing limit amount calculation means 2
5 is added to form a parallel image processing device.

As described above, according to the ninth embodiment, the device of the fourth embodiment is realized by combining the function expansions of the fifth to eighth embodiments. Therefore, there is an effect that any function expanding means can be selected from the function expansions of the fifth to eighth embodiments and the respective advantages can be used in combination.

[Embodiment 10] 19 is a block diagram showing a parallel image processing apparatus according to Embodiment 10 of the present invention. In FIG. 19, reference numeral 26 is a unit area division correspondence type image area allocating means which is an extension of the function of the image area allocating means 2 in FIG. Reference numeral 27 is a unit area division corresponding type area division setting means in which the function of the area division method setting means 6 in FIG. 1 is expanded, and 28 is a unit area division corresponding type processing limit amount in which the function of the processing limit amount calculating means 7 in FIG. 1 is a unit area division correspondence type processing load calculation means in which the function of the processing load calculation means 8 in FIG. 1 is expanded, and 30 is a unit area division correspondence type image processing in which the function of the image processing execution means 9 in FIG. 1 is expanded. It is an execution means. Other configurations are the same as those in FIG. 20 is a flow chart showing the processing of the unit area division corresponding image area allocating means, FIG. 21 is a flow chart showing the processing of the unit area division corresponding processing load calculating means, FIG.
FIG. 7 is an explanatory diagram showing an image of division of processing in units of rows.

Next, the operation will be described. In the tenth embodiment, in order to divide and re-allocate the processing in the unit area, the unit area division corresponding image area allocating means 26, the unit area division corresponding area division setting means 27, the unit area division corresponding processing. The operations of the limit amount calculating means 28, the unit area division corresponding processing load calculating means 29, and the unit area division corresponding image processing executing means 30 are the image area allocating means 2 of the first embodiment,
It is different from the area division method setting means 6, the processing limit amount calculating means 7, the processing load calculating means 8, and the image processing executing means 9. However, the maximum processing amount per unit area setting unit 3, the used processor number setting unit 4, and the processed image information setting unit 5 operate in the same manner as in the first embodiment.

Image area allocating means 26 corresponding to unit area division
The operation will be described with reference to FIG. First, the initialization is basically the same as the image area allocation unit 2 of the first embodiment (step ST41). However, it differs from the first embodiment in that the method of allocating a processing area instructed by the unit area division-compatible area division setting means 27 is a method that takes line division into consideration. In this example, for the sake of brevity, the line-divided and reallocated lines will be treated as unprocessed lines. Here, if there is a line that has been line-divided and reallocated, the "remaining processing area" is incremented by 1 for this line.

After the initialization, the unit area division correspondence type image area allocating means 26 enters the "request waiting" state (step ST
42). In the "waiting for request" state, the operation when the "next processing" is requested from the unit area division corresponding processing load calculating means 29 will be described. Here, first, the "number of remaining processing areas" is checked (step ST43). When the "remaining processing area number" is larger than 0, the processing area allocation operation is performed. This procedure is basically the same as the image area allocating means 2 of the first embodiment, but the following two points are different. The first point is a method of determining the "next processing area". Here, the processing area allocation method instructed by the unit area division-compatible area division setting unit 27 is a method that considers line division.
Different from the first embodiment. For example, in the first embodiment,
After determining the size of the “next processing area”, areas having the management flag “not yet” corresponding to the “size of the next processing area” are sequentially selected from the position indicated by the “image allocation pointer”. On the other hand, in this example, first, it is checked whether there is an area in which the management flag is "minute" (step ST
44). If there is an area in which the management flag is "minute", the image area consisting of only one line is determined as the "next processing area". At this time, as the “next processing area” information, the number of scattered points at which the calculation after the reallocation is started (for example, 100 scattered points are originally found in one line, 70 of them are processed, and the remaining 30 are processed). When is reassigned, as the number of scattered points to start the calculation after reassignment,
71 is designated) is also added (step ST
45). Further, the "remaining processing area" is decreased by one. When there is no area in which the management flag is "minute", the "next processing area" is determined by the same method as in the first embodiment (steps ST46 and ST47). It should be noted that the management flag of "minute" is an area in which a line is divided and reassignment is requested. The second point is "process limit amount"
There is calculation and instructions. In the first embodiment, the processing limit calculation means 7 calculates the “processing restriction amount” on the assumption that the unit area is the minimum unit, and notifies the processing load calculation means 8 of this “processing restriction amount”. . On the other hand, in this example, the unit area division-compatible processing load calculating unit 29 adds to the “processing limit amount” (hereinafter, execution restriction) based on the assumption that the unit area is the minimum unit (row unit allocation). , A "processing restriction amount" (hereinafter, division constraint) serving as a determination criterion for allocation in which the unit area is divided (row processing is also divided) is calculated. The unit area division correspondence type image area allocating means 26 divides the execution constraint and the division constraint into two.
It is different from the first embodiment that the processing load calculation unit 29 corresponding to the unit area division is notified of the type of “processing limit amount”.

Next, in the unit area division correspondence type image area allocation means 26, when the "next processing" is requested from the unit area division correspondence type processing load calculation means 29, the operation when the "remaining processing area number" is 0 Indicates. Here, it operates in the same procedure as the image area allocating means 2 of the first embodiment (steps ST51 to ST51).
Step ST53).

Image region allocating means 26 corresponding to unit region division
In the "waiting for request" state, the operation in the case where "reallocation" is requested from the unit area division-compatible processing load calculating means 29 will be described. In this case, first, the "number of remaining processing areas" is updated (step ST54). Here, "reallocation"
In addition to the requested area (in units of rows), the number of rows for which processing is divided and “reassignment” is requested is added as the “remaining processing area number”. Therefore, if there is a "reallocation" due to line division, "the number of remaining processing areas" +
"Size of reallocation area (row unit: from the row next to the boundary row to the last row of the" next processing area ")" + 1 row "becomes the new" number of remaining processing areas ". Next, the management flag of the execution management table for each processing unit is updated for the area for which “reallocation” is requested. Here, the management flag is updated to "minutes" for the rows requested to be "reassigned" by dividing the rows, and the management flags for the other rows are updated to "not yet" (step ST5
5). In addition, regarding the line where the management flag is set to "minute",
The position at which the reallocation is started (estimated number of scattered points) is also recorded. The subsequent procedure, the setting when the image allocation pointer is "empty", the confirmation of the in-execution flag of each processor, the "restart of execution" notification, and the like are the same as those of the image area allocation unit 2 of the first embodiment ( Step ST56 to step ST58).

Processing load calculation means 29 corresponding to unit area division
The operation will be described with reference to FIG. The unit area division-compatible processing load calculation unit 29 receives an instruction to start execution from the unit area division-compatible image area allocation unit 26, requests the next process, and waits for the reply (step ST61).
~ Step ST69). Here, when the answer is “temporary stop” and the subsequent processing procedure, it is the same as that of the processing load calculating means 8 of the first embodiment (step ST84, step ST85).

When the response to the request for the next processing is the notification of the information of the "next processing area" and the "processing limit amount", it is checked whether or not the "next processing area" is the reallocation of the line-divided range ( Step ST64). Here, in the case of reallocation of the line-divided range, since the information of the position to start the calculation after the reallocation is added as the "next processing area" information, the judgment is made based on this. When the "next processing area" is not the reallocation of the line-divided range, first, the "boundary row" is obtained by the same procedure as the processing load calculation means 8 of the first embodiment. However, here, the unit area division correspondence type image area allocating means 2
The first embodiment is that two types of constraint conditions, that is, an execution constraint and a division constraint, are set as the "processing limit amount" from 6 above.
Different from Here, using execution constraints, the first embodiment
The "boundary line" is obtained by the same procedure as in step ST65.
-Step ST70). After the "boundary line" is determined, it is determined whether or not to divide the line-by-line process using the division constraint of the "processing limit amount" (step ST71). This constraint condition is that the estimated number of scattered points in the "boundary line" is a certain number or more, or the total number of estimated scattered points in the "next processing area" assigned this time is a certain number or more. Conditions such as can be set arbitrarily. Where "boundary line"
If it is determined that the process division for each line is necessary, the "boundary line" is divided into the "super-resolution execution region" and the "unprocessed region". In this division, the number of estimated scattering points specified by the division constraint from the beginning is set in the "super-resolution execution area", and the estimated scattering points not included in the "super-resolution execution area" are set in the "unprocessed area". (Step ST72, step ST73). In addition, information is also recorded that the estimated scattering point at the beginning of the “unprocessed area” is the number from the beginning. On the other hand, if it is determined that it is not necessary to divide the line-by-line processing in "boundary line",
The entire "boundary line" is set to the "super-resolution execution area" (step ST74). Next, in the unit area division-compatible processing load calculating means 29, from the head of the "next processing area" to "boundary line"
Up to the line before is additionally set in the "super-resolution execution area" (step ST75). Further, when the "boundary line" is not the final line, the region from the line next to the "boundary line" to the last line of the "next processing region" is additionally set to the "unprocessed region" (step ST73).

After the setting of the "super-resolution execution area" and the "unprocessed area" is completed, the unit area division corresponding processing load calculation means 29 allocates the "unprocessed area" to the unit area division corresponding image area allocation. A procedure for requesting reallocation to the means 26 (step ST81) and execution of super-resolution processing of the "super-resolution execution area" are instructed to the unit area division correspondence type image processing execution means 30 (step ST82), and next. The procedure for returning to the processing request is basically the same as that of the processing load calculation means 8 of the first embodiment. However, the following two points are different from the first embodiment. First, there is information added in the re-allocation request to the unit area division compatible image area allocating means 26 as a first point. In the tenth embodiment, when a reallocation request is made to the unit area division correspondence type image area allocating means 26, “a reallocation part in a row unit” is executed.
And information is transmitted in a format that distinguishes between the "reassigned part that divides the line". In addition, regarding the "reassigned portion that divides the line", information indicating how many estimated scattered points at the beginning of the "unprocessed area" are added is also notified. If there is no "reallocation part for dividing the line", only the information of the "reallocation part for each row" is notified. In this case, the same information as in the first embodiment is used. Next, as a second point, there is a range for instructing the image processing executing means 30 corresponding to unit area division to execute the super-resolution processing of the “super-resolution execution area”. In the tenth embodiment, as the “super-resolution execution region”, a line for executing the entire line, a line to be divided into lines and a range thereof (start scatter point and end scatter point) are designated, and a unit region is specified. The division-compatible image processing execution means 30 is instructed to execute the super-resolution processing of the “super-resolution execution area”. The unit area division correspondence type image processing executing means 30 executes the same processing as the image processing executing means 9 of the first embodiment for a row instructed to execute the entire row. On the other hand, for a line to be divided into lines and executed, the super-resolution processing of the “super-resolution execution region” is executed only for the start scattering points and the end scattering points of the target row.

On the other hand, in the case of reallocation of a range in which the "next processing area" is line-divided, the unit area division-compatible processing load calculating means 29 operates as follows. In this case, as the "next processing area" information, information on the position to start the calculation after the reallocation (from which scattering point the processing target starts) is added. Here, first, all the estimated scattering points of the row designated by the “next processing area” are calculated by the same method as the processing load calculation means 8 of the first embodiment. After that, among the obtained estimated scattering points, the estimated scattering points after the position where the calculation is started are set as “line division reprocessing target estimated scattering points” (step ST77). Next, "Estimated scattered points for line-break reprocessing"
Is determined by using the division constraint of the "processing limit amount" for whether or not the line-by-line processing is divided again (step ST7).
8). Here, if the division is not necessary, all the "line division reprocessing target estimated scattering points" are designated as the "super-resolution execution area" (step ST79). On the other hand, when it is necessary to perform division, the "super-resolution execution area" and "unprocessed area" can be divided into "super-resolution execution areas" using the "processing restriction amount" division restriction, as in the case of the above "boundary line" division. Divide (step ST80). However, the head of the "super-resolution execution area" here is the position where the calculation after the reallocation is started, and the specified number is selected from this position. In addition, for the estimated scatter point indicating the beginning of the “unprocessed area”, information about the number of rows from the beginning of the row is set as in the case of dividing the “boundary row” (here, the calculation after reassignment is performed). Not counting from the starting position). "Super-resolution execution area"
After the setting of the "unprocessed area" and the "unprocessed area", it is the same as the case where the "next processing area" is not the reallocation of the line-divided range. However, in the “super-resolution execution area” instructed to the unit area division-compatible image processing execution means 30, the start point of the line to be divided into lines and executed is not the head of the line but the unit area division-compatible image region. It is at the "position to start the calculation after the reallocation" instructed by the allocating means 26.

As described above, according to the tenth embodiment, when the processing load on a certain processor increases, it is possible to divide the unit area and reallocate the processing to another processor. As a result, with an image processing algorithm whose processing load changes depending on the input image, even if the load on a certain unit area increases significantly, the processing can be efficiently distributed to the processors that execute in parallel, and parallel processing with high execution efficiency can be performed. The image processing device can be obtained.

In the tenth embodiment described above, the division criterion in the unit area is shown by a simple method for the sake of simple explanation. However, this constraint condition is an element that determines the processing load of the unit area. In the example, an arbitrary selection method can be adopted as long as it is based on the estimated number of scattering points), and the same effect as that of the tenth embodiment can be obtained. Further, it is of course possible to set a constraint condition in which other conditions (size of image processing, etc.) are combined, and the same effect as that of the tenth embodiment is achieved. The same applies to the calculation method of the "next processing area" and the control using the management flag in the execution management table, as long as it is a method that can be assigned without leaving an unprocessed area corresponding to division in a unit area. An arbitrary “next processing area” calculation method or area management method can be adopted, and the same effect as that of the tenth embodiment can be obtained.

In the tenth embodiment described above, in order to explain the method in a concise manner, an example in which reassignment of a line-divided range is selected and executed preferentially has been described. It is also possible to mix and assign unprocessed rows (rows for which calculation has not been executed even once), and the same effect as in the tenth embodiment can be obtained.

Eleventh Embodiment FIG. 23 is a block diagram showing a parallel image processing apparatus according to Embodiment 11 of the present invention. In the figure, 31 is a unit area division availability condition setting means. Other configurations are the same as those in FIG.

Next, the operation will be described. In the eleventh embodiment, the unit area division-compatible processing limit amount calculation means 2
It is assumed that 8 is set as a criterion for the division restriction of the processing restriction amount, which is a condition in which the remaining processing amount of the entire image processing and the number of processors suspended for execution are combined. In this case, the unit area division corresponding processing load calculation unit 29 sets the combination condition calculated by the unit area division corresponding image area allocation unit 26 by the unit area division corresponding image processing amount calculation unit 28 as the processing limit amount. It will be specified as a criterion for division restriction. The unit area division compatible processing load calculation unit 29 collects the information specified in the combination condition, and therefore the unit area division permission / inhibition condition setting unit 31.
To collect information. The unit area division permission / inhibition condition setting means 31 refers to the unit area division correspondence type area division setting means 27 when receiving an instruction, and determines the remaining processing amount (see "remaining processing amount") in the entire image processing and execution. Information such as the number of processors suspended for waiting (see the number of processors whose execution flag is "OFF") is collected and transmitted to the unit area division-compatible processing load calculation means 29. When determining whether or not line division is possible, the unit area division-compatible processing load calculation unit 29 refers to the information from the unit area division availability condition setting unit 31 and determines a combination condition. It should be noted that the present apparatus operates in the same manner as the apparatus of the tenth embodiment, except for the procedure for designating the combination condition and collecting the combination condition when determining the possibility of line division.

As described above, according to the eleventh embodiment, as the conditions for judging whether or not the unit area can be divided, the remaining processing amount of the entire image processing and the number of processors suspended for execution wait. You can use the execution status such as. As a result, it is possible to select a load distribution suitable for the current execution status, and it is possible to obtain a parallel image processing device with high execution efficiency.

In the eleventh embodiment, as the information to be referred to by the unit area division permission / inhibition condition setting means 31, the remaining processing amount of the entire image processing and the number of processors suspended for execution are designated. Any information can be collected as long as it is the information managed by the unit area division correspondence type area division setting means 27, and the same effect as that of the above-described eleventh embodiment can be obtained. Further, with respect to the information acquired and used by the unit area division correspondence type area division setting means 27 such as the information of the processing target image held by the processing image information setting means 5, any information can be similarly collected. This is possible, and the same effect as the eleventh embodiment can be obtained.

Twelfth Embodiment 24 is a configuration diagram showing a parallel image processing device according to a twelfth embodiment of the present invention, in which reference numeral 32 is a pre-calculation result holding means.
Other configurations are the same as those in FIG.

Next, the operation will be described. The unit-area-division processing load calculation means 29 is a means that can hold the calculation result regarding the estimated scattering point for any row. It is also a means that can extract the calculation result regarding the estimated scattering point for the specified row from the saved calculation results. In the unit area division-compatible processing load calculation means 29 of the twelfth embodiment, when it is judged that a certain row needs to be divided for the first time, the calculation result regarding the estimated scattering point is held in the pre-calculation result holding means 32. Save to. Also, the unit area division-compatible image area allocation means 2
In the case of reassignment of the range in which the "next processing region" is divided into rows from No. 6, the information held by the pre-calculation result holding means 32 is used without re-executing the calculation for obtaining the estimated scattering points. The pre-calculation result holding means 32 is basically mounted in a common part between the processors. For example, SMP
If it is a computer, it is possible to implement it so that the pre-calculation result is held in the shared memory part.

As described above, according to the twelfth embodiment, when the unit area is divided and executed, the processor which first calculates the unit area to be divided records the calculation result of the preprocessing for the area. I'll do it. In the processor instructed to execute the other part of the unit area that was divided later,
By using this recorded calculation result, it is not necessary to calculate the pre-process again. Therefore, the calculation of the preprocessing can be saved in time, and there is an effect that a parallel image processing device with high execution efficiency can be obtained.

Thirteenth Embodiment FIG. 25 is a block diagram showing a parallel image processing device according to a thirteenth embodiment of the present invention. In the figure, 21 is cache information setting means, 33.
Is a unit area division cost calculation means. Other configurations are the same as those in FIG.

Next, the operation will be described. In the thirteenth embodiment, the unit area division-compatible processing limit amount calculation means 2
It is assumed that 8 sets the maximum and minimum estimated number of estimated scattering points that can be selected as a criterion for the division restriction of the processing restriction amount. In this case, the unit area division-compatible processing load calculation unit 29 instructs the unit area division cost calculation unit 33 to calculate the execution cost within the selectable maximum and minimum estimated scattering point ranges. In the unit area division cost calculation means 33, reference conditions for calculating the processing cost due to cache access in the calculation of the execution target are set. The unit area division cost calculation means 33 is the cache information setting means 21.
The cache miss penalty of the cache hit rate is obtained from the cache information of 1, and information such as the size of the image to be processed and the pixel size is obtained from the processed image information setting means 5.
The extent to which the cache access in the target image is advantageous or disadvantageous and its extent are roughly estimated. The unit area division cost calculation unit 33 calculates the cost from the position on the image where the estimated scattered points are processed when the execution cost calculation instruction is given from the unit area division corresponding processing load calculation unit 29. For example, where the estimated scattering points are dense, the cost is low because the data in the same cache can be used repeatedly. Get higher In the unit area division cost calculation means 33, the unit area division correspondence processing load calculation means 29 has an optimum number within the range of the designated maximum and minimum number of estimated scattering points according to the set processing cost reference conditions. The number of estimated scattering points is selected, and the unit area division correspondence processing load calculation means 29 is notified. The unit area division corresponding processing load calculation unit 29 determines to divide the processing of the target row by the number of estimated scattering points notified from the unit area division cost calculation unit 33. In addition,
The present apparatus operates in the same manner as the apparatus of the tenth embodiment, except for the procedure for designating the combination condition and collecting the same when determining whether or not line division is possible.

As described above, according to the thirteenth embodiment, the position where the unit area is divided can be determined in consideration of the cache access efficiency. As a result, there is an effect that the calculation in the area after the division can be executed in an efficient range, and a parallel image processing apparatus with high execution efficiency can be obtained.

Fourteenth Embodiment FIG. 26 is a block diagram showing a parallel image processing device according to the fourteenth embodiment of the present invention. In the figure, the configuration shown in FIG.
The pre-calculation result holding means 32 and the unit area division cost calculation means 33 are added to constitute a parallel image processing device.

As described above, according to the fourteenth embodiment, the device of the tenth embodiment is realized by combining the function expansions of the eleventh to thirteenth embodiments.
Therefore, there is an effect that any function expanding means can be selected from the function expandings of the eleventh to thirteenth embodiments and the respective advantages can be used in combination.

Fifteenth Embodiment 27 is a configuration diagram showing a parallel image processing device according to a fifteenth embodiment of the present invention. In the figure, 34 is load prediction setting means, and 35 is FIG.
It is a load prediction type area division method means in which the function of the area division method setting means 6 in FIG. Other configurations are the same as those in FIG.

Next, the operation will be described. When processes are distributed and executed by dynamic scheduling, if a heavy load appears in the latter half of the process (especially immediately before the end), it is difficult to evenly finish the processes in each processor, which is disadvantageous in parallel execution. become. This is a Guided-Self
The same tendency applies not only to the dynamic division method of general processing such as scheduling and Self scheduling, but also to the processing of the first to fourteenth embodiments.

Load prediction setting means 3 of the fifteenth embodiment
In 4, it is possible to set the load prediction for the image to be processed. For example, when a user who performs super-resolution processing in MUSIC visually sees an image to be processed and predicts that the processing load will be high in a local area of the image, the processing load in that area will increase. It is possible to set prediction information that is high and the processing load of other parts is low. As this setting method, it is possible to use a graphical user interface to specify on the screen displaying the image processing area. The load prediction type area division method unit 35 preferentially allocates to each processor from the processing area (row) predicted to have a high load, based on the prediction value designated by the load prediction setting unit 34. It is possible to select a different area division method.

In the present apparatus, the target image processing is executed in parallel by using the area division method which preferentially processes the processing area selected in the above procedure and predicted to have a high load. In addition, the parallel execution method of image processing here is
The apparatus is the same as the apparatus of the first embodiment except that the method of selecting the processing area is different.

As described above, according to the fifteenth embodiment, it is possible to give priority to an image region predicted to have a high load and allocate it to each processor based on the input predicted value. Therefore, it is possible to suppress the possibility that a high-load portion appears in the second half of the processing (especially immediately before the end), and the processing ends in each processor can be more evenly distributed (the load can be evenly distributed). There are advantages. Also,
By using the predicted value of the processing load, there is an advantage that the probability that a more appropriate image processing target area allocation method can be selected becomes high.

In the above-mentioned fifteenth embodiment, the method of designating the predicted value by the user has been described, but this designating method is arbitrary, and even if the designation is made by using another method or means, the above-mentioned embodiment is carried out. The same effect as that of the fifteenth aspect can be obtained. In the fifteenth embodiment, the load is unbalanced in the entire image processing area, and an area in which the load is large appears in the latter half of the image processing execution, and the load distribution is prevented from becoming difficult. On the contrary, in the present apparatus, when almost uniform load is predicted, a simple and low overhead processing allocation method can be selected. As described above, in the present apparatus, not only the case corresponding to the heavy load part, but also various types of this method can be selected based on the input prediction information. The same effect as that of the fifteenth aspect can be obtained.

Sixteenth Embodiment 28 is a block diagram showing a parallel image processing apparatus according to Embodiment 16 of the present invention. In the figure, 36 is a brightness value calculation formula setting means, 37 is a brightness value calculation execution means, and 38 is a pixel reference load prediction calculation means. Is. Other configurations are the same as those in FIG.

Next, the operation will be described. The brightness value calculation formula setting means 36 sets a calculation formula for calculating the brightness value for each pixel in the image processing area to be processed. For example, the calculation formula of the brightness value for the complex image area to be processed is “√
((Square of real part) + (square of imaginary part)) ”is designated. The brightness value calculation executing means 37 calculates the brightness value for each pixel in the image processing area to be processed according to the calculation formula specified by the brightness value calculation formula setting means 36. The pixel reference load prediction calculation means 38 is the brightness value calculation execution means 37.
The predicted value of the processing load for each unit area is calculated based on the calculation result of. For example, the pixel reference load prediction calculation means 3
In 8, the predicted value of the processing load is calculated in proportion to the sum of the brightness values in the unit area (in this example, the sum of the brightness values in each row is obtained, and the estimated number of scattering points is calculated according to the size of the sum. Predict the number of). Further, the pixel reference load prediction calculation means 38 sets the calculated prediction value in the load prediction setting means 34. The load prediction type area division method unit 35 determines the area division method based on the prediction value set in the load prediction setting unit 34, as in the fifteenth embodiment.

As described above, according to the sixteenth embodiment, the processing load can be predicted from the information of each pixel of the image to be processed. Thereby, there is an advantage that the load prediction accuracy can be increased in a case where the load prediction accuracy from the information of each pixel is high. Improving the accuracy of prediction has the advantage of being able to select a more appropriate image processing target area allocation method.

In the sixteenth embodiment, “√ ((square of real part) + (square of imaginary part))” is used as the formula for calculating the brightness value.
However, it is possible to set an arbitrary calculation formula in the brightness value calculation formula setting means 36.
The brightness value calculation executing means 37 can calculate the brightness value corresponding to the specified calculation formula, and can achieve the same effect as that of the above-described sixteenth embodiment.

Seventeenth Embodiment 29 is a block diagram showing a parallel image processing device according to a seventeenth embodiment of the present invention. In FIG. 29, 39 is a pixel reference load prediction condition setting means, and 40 is a pixel reference load prediction calculation means 3 in FIG.
The pixel reference load prediction calculation means corresponding to the measurement condition, which is obtained by expanding the function of 8. Other configurations are the same as those in FIG. 28.

Next, the operation will be described. In the seventeenth embodiment, the pixel reference load prediction condition setting means 39 can set the calculation method of the prediction load for the brightness value of each pixel. Further, the measurement condition corresponding pixel reference load prediction calculation means 4
With 0, the predicted load for each unit area can be calculated in accordance with the predicted load calculation method designated by the pixel reference load prediction condition setting means 39.

For example, the pixel reference load prediction condition setting means 3
It is assumed that “the number of pixels having a brightness value exceeding a certain“ threshold value ”for each unit area is specified as the calculation method of the predicted load in 9. At this time, the measurement condition-corresponding pixel standard load prediction calculation is performed. The means 40 first instructs the brightness value calculation executing means 37 to calculate the brightness value at each pixel, where the brightness value calculation formula corresponding to this load prediction is the brightness value as in the sixteenth embodiment. The brightness value calculation executing means 37 is set in the calculation formula setting means 36, and the brightness value of each pixel is calculated according to the setting. After the brightness value calculation in 37 is completed, the number of pixels having a brightness value exceeding the “threshold value” specified by the pixel reference load prediction condition setting unit 39 for each unit area (each row in this example). And load this as the predicted load That. The measurement condition corresponding pixel reference load prediction calculation means 40 sets the prediction value calculated here in the load prediction setting means 34. Thereafter, the load prediction type area division method means 35 determines the area division method based on the predicted value set in the load prediction setting means 34 in the same procedure as in the fifteenth embodiment.

As described above, according to the seventeenth embodiment, the calculation formula for predicting the processing load is set based on the information of each pixel of the image to be processed, and the unit is calculated using the calculation formula. The predicted load for each area can be calculated. Thereby, there is an advantage that the calculation method of the load prediction from the information of each pixel can be flexibly set. Further, it is expected that the load prediction accuracy can be improved by selecting a prediction load calculation method from the knowledge of the processing target image and the preconditions.

In the above-mentioned seventeenth embodiment, the method of calculating the predicted load is described by using the number of pixels exceeding the "threshold" for the luminance value as a reference, but here, a method of calculating an arbitrary predicted load is used. It can be set, and the first embodiment
The same effect as that of No. 7 can be obtained. For example, a calculation method based on the brightness value of the pixel having the maximum (peak) value for each unit area, a calculation method based on the number of pixels included in a certain brightness value range, and a calculation method combining these conditions. Etc. can be set, and the same effect as that of the above-mentioned seventeenth embodiment can be obtained.

Eighteenth Embodiment 30 is a block diagram showing a parallel image processing apparatus according to Embodiment 18 of the present invention. In the figure, 41 is a characteristic region extraction setting unit, 42 is a characteristic region extraction unit, and 43 is a pixel reference load prediction calculation in FIG. It is a characteristic region load prediction calculation means that extends the function of the means 38. Other configurations are the same as those in FIG. 28.

Next, the operation will be described. In the eighteenth embodiment, the characteristic region extraction setting unit 41 can set the method of calculating the characteristic region serving as the reference of load prediction from the luminance value of the pixel. The characteristic region extraction means 42 can extract the characteristic region according to the setting of the characteristic region extraction setting means 41. The characteristic region load prediction calculation unit 43 can calculate the predicted load for each unit region from the characteristic region extracted by the characteristic region extraction unit 42.

For example, it is assumed that "a continuous region of pixels whose luminance value exceeds the" threshold value "within the unit region" is set in the characteristic region extraction setting means 41 as the characteristic region calculating method.
The characteristic region extracting unit 42 extracts, as the characteristic region, a continuous region of pixels whose luminance value exceeds the “threshold value” for each unit region (for each row in this example) according to the setting of the characteristic region extraction setting unit 41. To do. Here, as in the sixteenth embodiment, the brightness value calculation formula corresponding to this load prediction is set in the brightness value calculation formula setting means 36, and the brightness value calculation executing means 37.
Calculates the brightness value at each pixel according to the setting. The characteristic region extracting means 42 extracts a characteristic region based on the brightness value calculated by the brightness value calculation executing means 37. It is assumed that the characteristic region load prediction / calculation unit 43 in this example is set to calculate the predicted load for each unit region based on the maximum size of the characteristic region. The characteristic region load prediction calculation unit 43 selects the characteristic region having the maximum size for each unit region from the characteristic regions extracted by the characteristic region extraction unit 42, and determines the predicted load based on the size. In the characteristic region load prediction calculation means 43, the prediction value calculated here is used as the load prediction setting means 3
Set to 4. After this, in the same procedure as in the fifteenth embodiment, based on the predicted value set in the load prediction setting means 34,
The load prediction type area division method means 35 determines the area division method.

As described above, according to the eighteenth embodiment, it is possible to set a characteristic region as a reference of load prediction from the brightness value of a pixel and predict the processing load from the information of this characteristic region. The advantage of increasing the probability of improving the load prediction accuracy by making it possible to predict the processing load not only on a pixel-by-pixel basis but also on the basis of a characteristic region that is a local set calculated based on the brightness value of pixels There is. Improving the accuracy of prediction has the advantage of being able to select a more appropriate image processing target area allocation method.

In the eighteenth embodiment, the method of calculating the characteristic region has been described by taking an example of "a continuous region of pixels whose luminance value exceeds the" threshold value "in the unit region", but here it is arbitrary. The method of calculating the characteristic region can be set, and the same effect as that of the above-described eighteenth embodiment can be obtained.

Nineteenth Embodiment 31 is a block diagram showing a parallel image processing device according to a nineteenth embodiment of the present invention, in which reference numeral 44 denotes a characteristic region load prediction condition setting means,
Reference numeral 45 denotes a characteristic condition load predictive calculation means corresponding to the measurement condition, which is obtained by expanding the function of the characteristic area load predictive calculation means 43 in FIG. Other configurations are the same as those in FIG.

Next, the operation will be described. In the nineteenth embodiment, the characteristic region load prediction condition setting means 44 can set the calculation method of the predicted load for the characteristic region. The measurement condition corresponding characteristic region load prediction calculation means 45 can calculate the predicted load for each unit area according to the setting of the characteristic region load prediction condition setting means 44.

For example, the characteristic region load prediction condition setting means 44
In addition, it is assumed that “the number of feature areas larger than a certain size in the unit area” is set as the calculation method of the predicted load. At this time, the characteristic condition load prediction calculation means 4 corresponding to the measurement condition
5 first instructs the characteristic region extracting means 42 to extract a characteristic region. The method of extracting the characteristic region by the characteristic region extracting means 42 is the same as in the eighteenth embodiment. Next, the measurement condition corresponding characteristic region load prediction calculation means 45 is the characteristic region extracted by the characteristic region extraction means 42 for each unit region, and is larger than the size set by the characteristic region load prediction condition setting means 44. The number is counted and calculated as a predicted value. The predicted value calculated by the characteristic condition load prediction calculation means 45 corresponding to the measurement condition is set in the load prediction setting means 34, and based on this predicted value, the load prediction type area division method means in the same procedure as in the fifteenth embodiment. 35 determines the area division method.

As described above, according to the nineteenth embodiment, a calculation region for setting a characteristic region serving as a reference for load prediction from the brightness value of a pixel and predicting the processing load for the information of this characteristic region. Can be set, and the predicted load for each unit area can be calculated using the calculation formula. As a result, not only in pixel units,
There is an advantage that the calculation method of the load prediction based on the information of the characteristic region, which is a local set calculated based on the brightness value of the pixel, can be flexibly set. Further, it is expected that the load prediction accuracy is improved by selecting a method for calculating the predicted load from the knowledge of the image to be processed and the preconditions.

In the nineteenth embodiment, the method of calculating the predicted load is described as "the number of feature areas larger than a certain size in a unit area", but here, the calculation of an arbitrary predicted load is performed. The method can be set, and the same effect as that of the nineteenth embodiment can be obtained. For example,
“The maximum (peak) value of the brightness values in the feature area is the“ threshold ”
The number of objects exceeding ",""sum of brightness values in the feature area",
“Average value of brightness values in feature area”, “number of feature areas exceeding a certain size”, conditions combining these, and the like can be set, and the same effect as in the nineteenth embodiment can be obtained. it can.

Embodiment 20. 32 is a block diagram showing a parallel image processing device according to a twentieth embodiment of the present invention. In the figure, reference numeral 46 is a reference area setting means, and 47 is FIG.
8 is a reference area load prediction calculation means in which the function of the pixel reference load prediction calculation means 38 in FIG. Other configurations are the same as those in FIG. 28. FIG. 33 is an explanatory diagram showing the reference area.

Next, the operation will be described. In the twentieth embodiment, the reference area setting means 46 sets an image area within a certain range as the reference area. The reference area load prediction calculation means 47 calculates the predicted load from the brightness value in the reference area set by the reference area setting means 46.

For example, the reference area setting means 46 is shown in FIG.
It is assumed that a certain pixel and 8 pixels adjacent to the certain pixel are set as the reference area as shown in FIG. Further, it is assumed that the reference area load prediction calculation means 47 is set to calculate the predicted load as follows. First, the average of the brightness values with respect to the reference region is obtained for each pixel. Next, the sum of the brightness values averaged for each pixel in the unit area is obtained. Finally, the predicted load in the unit area is determined from this total. At this time, in the reference area load prediction calculation means 47, first, the reference area setting means 4
According to the reference area set by 6, the average of the brightness values in the reference area for each pixel is obtained. The method of calculating the brightness value is the same as that in the sixteenth embodiment. Next, for each unit area, the total sum of the brightness values averaged for each pixel is calculated and used as the predicted load in that unit area. Finally, the reference area load prediction calculation unit 47 sets the calculated predicted load in the load prediction setting unit 34. Based on the predicted load set in the load prediction setting means 34, the load prediction type area division method means 35 determines the area division method in the same procedure as in the fifteenth embodiment.

As described above, according to the twentieth embodiment, it is possible to set an image area in a fixed range as a reference area and predict the processing load from the information of the brightness value in this range. By predicting the load based on the information calculated based on the brightness value from the image area of the fixed fixed range, there is an advantage that the probability of improving the load prediction accuracy increases. Improving the accuracy of prediction has the advantage of being able to select a more appropriate image processing target area allocation method.

In the twentieth embodiment, an example in which the reference region is a region as shown in FIG. 33 has been described. However, a reference region of a shape and a range can be set here, and the same as in the twentieth embodiment. The effect of can be produced.

Embodiment 21. 34 is a block diagram showing a parallel image processing apparatus according to Embodiment 21 of the present invention. In the figure, 48 is a reference area load prediction condition setting means, and 49 is a reference area load prediction calculation means 4 in FIG.
It is a reference condition load prediction calculation means corresponding to the measurement condition, which is an expanded function of 7. Other configurations are the same as those in FIG.

Next, the operation will be described. In the twenty-first embodiment, the reference area load prediction condition setting means 48 can set the calculation method of the prediction load based on the brightness value in the reference area. In addition, the measurement condition compatible reference area load prediction calculation means 4
9 can calculate the predicted load for each unit area from the predicted load calculation method set by the reference area load prediction condition setting means 48.

For example, the reference area load prediction condition setting means 4
It is assumed that the following method of calculating the predicted load is designated for the same reference region as in the twentieth embodiment. First, pixels in which the total sum of brightness values in the reference region is larger than "threshold value 1" are extracted, and this set is referred to as "pixel group 1". Next, in "pixel group 1", the sum of the differences between the brightness value of the target pixel and the brightness values of the other pixels in the reference area is calculated. A pixel whose total sum is smaller than "threshold value 2" is extracted and set as "pixel group 2". Finally, the number of pixels belonging to the “pixel group 2” is counted for each unit area, and this is set as the predicted load value. The measurement condition corresponding reference area load prediction calculation means 49 advances the processing according to the method of calculating the predicted load set by the reference area load prediction condition setting means 48. In this example, first, in accordance with the reference area set by the reference area setting means 46, pixels whose sum of brightness values in the reference area is larger than “threshold value 1” are extracted,
This set is called “pixel group 1”. The method of setting the reference area by the reference area setting means 46 is the same as in the twentieth embodiment. Next, in the measurement condition corresponding reference area load prediction calculation means 49, for the pixels included in the “pixel group 1”,
The sum of the differences between the brightness value of the target pixel and the brightness values of the other pixels in the reference area is calculated. After that, a pixel whose total sum is smaller than "threshold value 2" is extracted and set as "pixel group 2".
After extracting the “pixel group 2”, the measurement condition corresponding reference area load prediction calculation unit 49 counts the number of pixels belonging to the “pixel group 2” for each unit area, and sets this as the value of the predicted load.
The measurement condition-corresponding reference area load prediction calculation means 49 sets the value of the predicted load calculated in the above procedure in the load prediction setting means 34. Based on the predicted load set in the load prediction setting means 34, the load prediction type area division method means 35 determines the area division method in the same procedure as in the fifteenth embodiment.

As described above, according to the twenty-first embodiment, an image area in a certain range is set as a reference area, and a calculation formula for predicting a processing load is set for information in the reference area. The predicted load for each unit area can be calculated using the calculation formula. Thereby, there is an advantage that the calculation method of the load prediction can be flexibly set by using the local information in the reference region having the fixed shape. It is also expected that the load prediction accuracy can be improved by, for example, selecting a predictive load calculation method from the knowledge of the image to be processed and the preconditions. The super-resolution processing used in the description of this twenty-first embodiment is processing for extracting buried scattering points. The processing load here is determined by the number of buried scattering point candidates. When predicting the load of such processing, it is possible to improve the prediction accuracy by determining various prediction calculation formulas on the basis of a change in a certain image area as in the device of the twenty-first embodiment. There is a high advantage.

In the twenty-first embodiment, as the method of calculating the predicted load, the method of obtaining the predicted load by a series of calculations has been described, but any calculation method of the predicted load can be set here. The same effect as that of the embodiment 21 can be obtained. For example, “The peak value of the luminance value for each pixel is calculated for each pixel, and in the unit area,
It is also possible to make a setting such that “the sum of the peak values for each pixel is used as the prediction load”, and the same effect as that of the twenty-first embodiment can be obtained.

Twenty-second Embodiment. 35 is a block diagram showing a parallel image processing apparatus according to Embodiment 22 of the present invention. In FIG. 35, reference numeral 50 is an inclination reference setting means, and 51 is FIG.
8 is a tilt load prediction calculation means that extends the function of the pixel reference load prediction calculation means 38 in FIG. Other configurations are the same as those in FIG. 28.

Next, the operation will be described. In the twenty-second embodiment, the tilt reference setting means 50 can set the reference tilt. The gradient load prediction calculation means 51 calculates the predicted load from the reference gradient set by the gradient reference setting means 50.

For example, the inclination reference setting means 50 sets the reference inclination to the amount of change between adjacent pixels on the "threshold value".
It is assumed that those that exceed (both increase and decrease) have been set. In addition, it is assumed that the slope load prediction calculation means 51 is set so that the number of slopes exceeding the “threshold” is used as the predicted value. At this time, the inclination load prediction calculation means 51 calculates the amount of change in the brightness value between each pixel in the unit area and its neighboring pixels. The method of calculating the brightness value is the same as that in the sixteenth embodiment. Next, the slope load predictive calculation means 51 counts those whose change amount exceeds the “threshold value” (both increase and decrease), and calculates the number as the predictive load for the unit area. Finally, the gradient load prediction calculation means 51 sets the calculated prediction load in the load prediction setting means 34. Based on the predicted load set in the load prediction setting means 34, the load prediction type area division method means 35 determines the area division method in the same procedure as in the fifteenth embodiment.

As described above, according to the twenty-second embodiment, it is possible to set the amount of change in the luminance value of a pixel as the slope and predict the processing load from the information of this amount of change. Predicting the load based on the amount of change has the advantage of increasing the probability of improving the load prediction accuracy. Improving the accuracy of prediction has the advantage of being able to select a more appropriate image processing target area allocation method.

In the twenty-second embodiment, an example in which the amount of change between adjacent pixels exceeds the "threshold value" has been described as the reference inclination, but here, the amount of change is calculated in the calculation range ( (Not only the pixels on both sides, but it is also possible to use three adjacent pixels on both sides) and the criterion for determining the amount of change can be set arbitrarily, and the same effect as in the above-described twenty-second embodiment can be obtained. .

Twenty-third Embodiment. 36 is a block diagram showing a parallel image processing device according to a twenty-third embodiment of the present invention. In FIG. 36, reference numeral 52 denotes inclination load prediction condition setting means, 5
Reference numeral 3 denotes a measurement condition-compatible tilt load prediction calculation means that extends the function of the tilt load prediction calculation means 51 in FIG.
Other configurations are the same as those in FIG.

Next, the operation will be described. In the twenty-third embodiment, the slope load prediction condition setting means 52 can set the calculation method of the prediction load for the slope. Further, the measurement condition corresponding tilt load prediction calculation means 53 can calculate the predicted load for each unit area by the calculation method set by the tilt load prediction condition setting means 52.

For example, the inclination load prediction condition setting means 52
In addition, it is assumed that, in the same gradient setting as in the twenty-second embodiment, a calculation method is specified in which the prediction load is "the number of pixels in a range in which three or more pixels satisfy the gradient condition continue" in the unit area. In this example, the measurement condition corresponding tilt load prediction calculation unit 53 calculates the predicted load in the following procedure according to the setting of the tilt load prediction condition setting unit 52. The measurement condition compatible tilt load prediction calculation means 53 firstly, the tilt reference setting means 5
Pixels that match the inclination condition of 0 are extracted (set 1). Here, the setting method of the inclination reference setting means 50 is the same as that of the twenty-second embodiment. Next, the measurement condition corresponding gradient load prediction calculation means 53 extracts the pixels included in the set 1 and the pixels on both sides thereof also included in the set 1 (referred to as a set 2). After extracting the set 2, the measurement condition corresponding gradient load prediction calculation means 53 counts the number of pixels belonging to the set 2 for each unit area and sets this as the value of the predicted load. The measurement condition-compatible tilt load prediction calculation means 53 sets the value of the predicted load calculated in the above procedure in the load prediction setting means 34. Based on the predicted load set in the load prediction setting means 34, the load prediction type area division method means 35 determines the area division method in the same procedure as in the fifteenth embodiment.

As described above, according to the twenty-third embodiment, the amount of change in the brightness value of a pixel is set as a slope, and a calculation formula for predicting the processing load from the information on the amount of change is set.
The predicted load for each unit area can be calculated using the calculation formula. Thereby, there is an advantage that the calculation method of the load prediction can be flexibly set by using the change amount of the brightness value of the pixel. Also, from the knowledge and prerequisites for the image to be processed,
It can also be expected to improve the load prediction accuracy by selecting a method for calculating the predicted load.

In the twenty-third embodiment, an example in which "the number of pixels in the range in which three or more pixels satisfy the slope condition are continued" has been described as the method for calculating the prediction load.
An arbitrary predictive load calculation method can be set, and the same effects as those of the above-described Embodiment 23 can be obtained.

Twenty-fourth Embodiment 37 is a block diagram showing a parallel image processing apparatus according to Embodiment 24 of the present invention. In the figure, 54 is a photographing area load prediction calculation means. Other configurations are the same as those in FIG.

Next, the operation will be described. In the twenty-fourth embodiment, the SAR image to be processed is often captured by a satellite or an aircraft. At this time, if the areas to be photographed are the same, the load can be predicted from the past image information and the like. The shooting area load prediction calculation unit 54 can record a shooting area and a processing load prediction for an image of the area. When the area in which the image to be processed is photographed is input, the photographing area load prediction calculation unit 54 extracts the prediction of the processing load for the image of the region and sets it in the load prediction setting unit 34.
Based on the predicted load set in the load prediction setting means 34, the load prediction type area division method means 35 determines the area division method in the same procedure as in the fifteenth embodiment.

As described above, according to the twenty-fourth embodiment, the processing load can be predicted from the information on the area where the image was taken. By predicting the load based on the information about the area where the image was captured, there is an advantage that the probability of improving the load prediction accuracy increases. Improving the accuracy of prediction has the advantage of being able to select a more appropriate image processing target area allocation method.

Twenty-fifth Embodiment 38 is a block diagram showing a parallel image processing device according to a twenty-fifth embodiment of the present invention. In the figure, the luminance value calculation formula setting means 3 is added to the configuration of FIG.
6, luminance value calculation executing means 37, pixel reference load prediction condition setting means 39, measurement condition corresponding pixel reference load prediction calculation means 40, characteristic area extraction setting means 41, characteristic area extraction means 42,
Characteristic area load prediction condition setting means 44, measurement condition corresponding characteristic area load prediction calculation means 45, reference area setting means 46, reference area load prediction condition setting means 48, measurement condition corresponding reference area load prediction calculation means 49, inclination reference The configuration includes a setting unit 50, a tilt load prediction condition setting unit 52, a measurement condition corresponding tilt load prediction calculation unit 53, and a photographing area load prediction calculation unit 54. 55 is a prediction calculation weight setting means. Other configurations are the same as those in FIG.

Next, the operation will be described. The twenty-fifth embodiment is realized by combining the apparatus of the sixteenth embodiment with the functional expansion of the seventeenth to twenty-fourth embodiments. Further, the prediction calculation weighting setting means 55 can be used for weighting the prediction values between the prediction means whose functions have been expanded according to the seventeenth to twenty-fourth embodiments.

As described above, according to the twenty-fifth embodiment, the predicting means whose functions are expanded according to the seventeenth to twenty-fourth embodiments can be combined and used flexibly, and the accuracy of prediction calculation can be improved. There is an effect that can be improved.

Twenty-sixth Embodiment 39 is a block diagram showing a parallel image processing apparatus according to Embodiment 26 of the present invention. In the figure, 56 is a predicted value recording means, 57 is a processing result recording means, and 58 is a prediction parameter correction means. Other configurations are the same as those in FIG. 37.

Next, the operation will be described. Here, the predicted value recording means 56 records the predicted value calculated in the past and the input condition. The processing result recording means 57 records the processing load calculated by the actual execution with respect to the past predicted value. The prediction parameter correction unit 58 corrects the prediction parameter such as the threshold value of the brightness value based on the prediction value of the prediction value recording unit 56 and the result of the actual processing time of the processing result recording unit 57.

For example, an example will be described in which a prediction value recording means 56, a processing result recording means 57, and a prediction parameter correction means 58 are added to the apparatus of the twenty-first embodiment. Here, the predicted value recording unit 56 records the predicted value calculated in the past by the reference area load prediction calculation unit 47 and set in the load prediction setting unit 34. Further, the actual execution time for the image processing executed based on this predicted value is the processing result recording means 5
7 records. The prediction parameter correction means 58 adjusts the values of “threshold value 1” and “threshold value 2” from the predicted value recorded by the predicted value recording means 56 and the actual processing load recorded by the processing result recording means 57. A value is calculated and the reference area load prediction condition setting means 48 is instructed to modify the calculation method of the predicted load. The reference area load prediction condition setting means 48 corrects the calculation method of the predicted load according to the instruction of the prediction parameter correction means 58. After that, the reference area load prediction condition setting means 48 uses the method of calculating the predicted load using the corrected threshold value.

As described above, according to the twenty-sixth embodiment, it is possible to correct the processing load prediction calculation method from the past processing load prediction and the actual processing load corresponding thereto. There is an advantage that the load prediction accuracy can be improved by correcting and improving the process load prediction calculation method from the past prediction result and the actual process load. There is also an advantage that a more appropriate image processing target area allocation method can be selected by improving the accuracy of prediction.

In the twenty-sixth embodiment, the twenty-first embodiment is described.
For the predicted value recording means 56 and the processing result recording means 57
, And an example in which the prediction parameter correction means 58 is added, but the configuration in which the above-described three means are added to any of the devices of the sixteenth to twenty-fifth embodiments is possible.
The same effect as that of the twenty-sixth embodiment can be obtained.

Twenty-seventh Embodiment 40 is a block diagram showing a parallel image processing device according to a twenty-seventh embodiment of the present invention. In FIG. Other configurations are the same as those in FIG. 37.

Next, the operation will be described. The image capturing condition load prediction correcting unit 59 records the image capturing conditions such as the weather, the image capturing date and time, the season, and the image capturing angle at the time of executing the previous predictive calculation, and the image capturing condition for the target image for which the new predictive calculation is performed. From, select the optimal prediction calculation correction.

For example, for the device of the twenty-sixth embodiment,
A parallel image processing apparatus is configured by adding the photographing condition load prediction correction means 59. Here, the photographing condition load prediction correction means 59
When the predicted value recording means 56 and the processing result recording means 57 respectively record the predicted value and the actual processing load, the corresponding conditions for shooting the image such as weather, shooting date, season, shooting angle, etc. are recorded. To do. Next, the photographing condition load prediction correction unit 59 selects a condition suitable for use in the new prediction from the recorded past photographing conditions when newly making a prediction, and a predicted value recording unit 56 for this condition is selected. From the predicted value and the actual processing load recorded by the processing result recording means 57,
The prediction parameter correction means 58 is instructed to calculate the correction method for the prediction value calculation. From here, the calculation method of the predicted load of the reference region load predicted condition setting means 48 is modified by the same procedure as in the twenty-sixth embodiment. Therefore, in the calculation of the new predicted value, the method of calculating the predicted load, which is modified in consideration of the shooting conditions, is adopted.

As described above, according to the twenty-seventh embodiment, when a new prediction is made from the conditions of the past photographing,
It is possible to correct the prediction calculation method of the processing load.
This has the advantage that the load prediction accuracy can be improved.
In the twenty-seventh embodiment, the SAR image to be processed is often captured by a satellite or an aircraft. At this time,
The SAR image obtained as a result is affected by the image capturing conditions such as the weather, the image capturing date and time, the season, the image capturing angle, and the like (varies even in the case of capturing the same place).
Due to the influence on the processing source image, the load in the super-resolution processing by MUSIC also changes. Therefore, it is desirable to consider the imaging condition also in the prediction calculation. The present apparatus can correct the prediction calculation method of the processing load by comparing the prediction calculation performed in the past with the actual processing amount based on the condition at the time of shooting. This is a predictive calculation in the case where the image to be processed is affected by the imaging conditions as in the twenty-seventh embodiment, and has the advantage that the prediction accuracy can be further improved.

In the twenty-seventh embodiment, an example has been described in which the predicted value and the actual measured processing load value corresponding to the condition suitable for new prediction are selected from the recorded past shooting conditions. From the above, even if a plurality of predicted values and actual measurement values of the processing load are selected and the selected values are weighted,
The same effect as that of the twenty-seventh embodiment can be obtained.

Embodiment 28. 41 is a block diagram showing a parallel image processing apparatus according to Embodiment 28 of the present invention. In the figure, 60 is an optimum prediction condition calculation means.
Other configurations are the same as those in FIG. 37.

Next, the operation will be described. The optimum prediction condition calculation means 60, when newly instructed to calculate the prediction value,
A similar condition is searched in the predicted value calculation input conditions recorded by the predicted value recording means 56, and the predicted value calculation condition suitable for this is selected. In addition, the parameter for the selected prediction value calculation condition is corrected from the prediction value for the selected prediction value calculation condition and the actual processing amount corresponding to the prediction result stored in the processing result recording unit 57.

For example, for the device of the twenty-sixth embodiment,
A parallel image processing apparatus is configured by adding the optimum prediction condition calculation means 60. In this example, the optimum prediction condition calculation means 60, when newly instructed to calculate the prediction value, the prediction value recording means 56.
From among the input conditions for the calculation of the predicted value recorded by, those that are close to the conditions specified by the new predicted value calculation are searched and one is selected. Next, the optimum prediction condition calculation means 60, the actual processing amount corresponding to the result of the calculated prediction value recorded in the prediction value recording means 56 for the selected condition and the prediction result held by the processing result recording means 57. And instruct the prediction parameter correction means 58 to correct the selected prediction calculation. Here, the prediction parameter correction means 58 performs the correction in the same procedure as in the twenty-sixth embodiment. After this correction,
A new predicted value is calculated using the selected corrected prediction calculation.

As described above, according to the twenty-eighth embodiment, when new prediction is performed, the prediction calculation executed under similar conditions can be selected from the input conditions of the past prediction calculations.
Furthermore, the calculation formula of the predicted value for this prediction condition can be corrected from the prediction result and the actual processing load amount for the selected prediction condition. This has the advantage that the load prediction accuracy can be improved.

In the above-mentioned twenty-eighth embodiment, an example of selecting one similar condition from past input conditions has been described.
Even when a plurality of conditions are selected and the selected values are weighted, the same effect as in the above-described twenty-eighth embodiment can be obtained.

Twenty-ninth Embodiment. 42 is a block diagram showing a parallel image processing apparatus according to Embodiment 29 of the present invention. In the figure, 61 is a pre-calculation selection setting means, 62 is a pre-calculation determining means, 63 is a pre-processing load amount calculating means, 6
Reference numeral 4 is a pre-processing compatible image area allocating means. Other configurations are the same as those in FIG.

Next, the operation will be described. The pre-calculation selection setting means 61 sets a judgment condition as to whether or not to perform the process by the pre-calculation. The pre-calculation determining unit 62 determines whether or not the process by the pre-calculation is performed based on the setting of the pre-calculation selection setting unit 61. The pre-processing load amount calculation unit 63, when instructed by the pre-calculation determination unit 62 to perform the process by the pre-calculation, calculates the processing load amount for each unit region in advance for the entire image region. Also, the result of the calculation for obtaining the processing load is held. Pre-processing compatible image area allocation means 64
Assigns processing areas such that the processing loads of the respective processors are equal based on the processing load amount calculated in advance by the preprocessing load amount calculating means 63 for each unit area.

For example, with respect to the apparatus of the first embodiment, a pre-calculation selection setting means 61, a pre-calculation judging means 62,
The pre-processing load amount calculation means 63 and the pre-processing compatible image area allocation means 64 are added to form a parallel image processing device. In this example, the pre-calculation selection setting means 61 is set to select the process by the pre-calculation when the maximum processing load amount for each unit area is 500 or more. The pre-calculation determination means 62 determines whether or not there is processing by pre-calculation based on the setting of the pre-calculation selection setting means 61 and the maximum processing load amount per unit area designated by the maximum processing amount per unit area setting means 3. .

Here, first, the processing procedure in the case where the maximum processing load is 1000 and the pre-calculation is selected at the time of the judgment by the pre-calculation judging means 62 will be shown. In this case, the pre-calculation determining unit 62 instructs the pre-processing load amount calculating unit 63 to start the pre-calculation. In the pre-processing load amount calculation means 63 instructed to start the pre-calculation, the entire image area is targeted,
The processing load amount for each unit area is calculated in advance. In this example, a calculation for obtaining an estimated scattering point is performed for each unit area (each row). In this calculation, the processing portion 12 that determines the processing load of the image processing program 11 as in the first embodiment.
To use. Further, the preprocessing load amount calculation means 63 records information on the calculated estimated scattering points. The calculation by the preprocessing load amount calculation means 63 may be realized in parallel. Since the calculation load here is constant for each unit area, it is easy to obtain a sufficient parallelization effect even if the calculation is implemented using a general parallel processing area division method.
After completion of the pre-calculation by the pre-processing load amount calculating means 63, the pre-processing corresponding image area allocating means 64 calculates the estimated scattered point number information for each row calculated by the pre-processing load amount calculating means 63 and the used processor number setting means. From the information on the number of processors to be used from 4, the processing area to be assigned to each processor is determined. The pre-processing compatible image area allocation unit 64 determines the processing area to be allocated to each processor so that the total number of estimated scattering points processed by each processor becomes equal. Here, since the allocation is based on the size of the processing that is known in advance, it is highly possible that the allocation can be evenly performed even if the method is implemented by using a general parallel processing area dividing method. Since the number of estimated scattering points assigned to the image processing executing means 9 of each processor has been adjusted to be equal,
Even if it is implemented by using a general area division method for parallel processing, it is easy to obtain a sufficient parallelization effect. After the processing area to be allocated to each processor is determined, the pre-processing compatible image area allocation means 64 instructs the image processing execution means 9 of each processor to execute the super-resolution processing for the processing area to be allocated. At this time, in the pre-processing compatible image area allocating means 64, the estimated scatter point corresponding to the allocated range is given to the image processing executing means 9 of each processor from the information of the estimated scatter point held by the pre-processing load amount calculating means 63. Pass the point calculation result. Here, the image processing execution means 9 of each processor executes the super-resolution processing in parallel, as in the first embodiment.

Next, at the time of the judgment by the pre-calculation judging means 62, the processing procedure when the maximum processing amount is 10 and the pre-calculation is not selected will be shown. In this case, the pre-calculation determination means 6
2 instructs the image area allocating means 2 to start processing by a normal calculation method. The image area allocating means 2 that has received the processing start instruction starts the calculation of the image processing algorithm in the same procedure as in the first embodiment.

As described above, according to the twenty-ninth embodiment, it is possible to select whether or not to use the processing method based on the pre-calculation and execute it. When the difference in the processing load dynamically determined for each unit area is particularly large, it may be difficult to cope with this by the dynamic scheduling that determines the allocation of the processing range at the time of execution. In the twenty-ninth embodiment, under such a possibility condition, the processing load is calculated in advance,
From this calculation result, the areas to be executed in parallel can be determined. On the contrary, under the condition that can be handled by the dynamic scheduling, the dynamic scheduling that is advantageous in the processing for each unit area is adopted and executed in parallel. As described above, the present apparatus has an advantage that an appropriate processing method can be selected according to the conditions under which the processing is performed.

In the twenty-ninth embodiment, the example in which the maximum processing amount for each unit area is 500 or more has been described as the determination condition for the pre-calculation. However, the determination condition is set for each unit area in the image processing algorithm. Any condition can be set as long as it is a condition that can determine whether or not the precalculation can be performed by estimating how much the difference between the maximum load and the minimum load (the difference in the processing time) is. The same effect as can be obtained.

In the twenty-ninth embodiment, compared with the first embodiment, the pre-calculation selection setting means 61, the pre-calculation judging means 62, the pre-processing load amount calculating means 63, and the pre-processing compatible image area allocating means 64. However, the configuration including the above-described four means can be added to any of the devices of the second to the twenty-eighth embodiments, and the same effect as that of the twenty-ninth embodiment can be obtained. Can be played. For example, in the case of adding to the third embodiment, it is possible to make a pre-computation determination based on the actually measured maximum processing time and minimum processing time for each unit area. There is an effect that it is possible to make an accurate judgment. In addition, in addition to the tenth embodiment, there is an effect that a parallel image processing device that also divides the unit area based on the result of the pre-calculation can be obtained.

Embodiment 30. 43 is a block diagram showing a parallel image processing apparatus according to Embodiment 30 of the present invention. In FIG. 43, reference numeral 65 denotes an image information reference type image area allocating means which is an extension of the function of the image area allocating means 2 in FIG.
Reference numeral 66 is an image information reference type processing load calculation means which is an extension of the function of the processing load calculation means 8 in FIG. 1, and 67 is an image information reference type area division method setting means which is an extension of the function of the area division method setting means 6 in FIG. is there. Other configurations are the same as those in FIG. FIG. 44 is a flowchart showing the processing of the image information reference type image area allocating means, FIG. 45 is a flowchart showing the processing of the image information reference type processing load calculating means, and FIG.
47 is a configuration diagram showing an implementation image of the parallel image processing device at the time of parallel execution, FIG. 47 is an explanatory diagram showing the concept of the operation of the execution management table, FIG. 48 is the “next process” assigned to the image information reference type processing load calculation means. 49 is an explanatory diagram showing an example of “super-resolution execution area” and “super-resolution unprocessed area” for “area”; FIG.
Is an explanatory diagram showing an implementation image when processing is performed with a four-processor configuration, and FIG. 50 is an explanatory diagram showing a concept of an allocation method when the image information reference type area division method setting means performs reallocation.

Next, the operation will be described. The operation of the image information reference type image area allocating means 65 will be described with reference to FIG. In the image information reference type image area allocating means 65,
First, necessary information for initialization is obtained from the maximum processing amount setting unit 3 per unit area, the used processor number setting unit 4, and the image information reference type area division method setting unit 67. From the used processor number setting means 4,
The maximum processing number per unit area is obtained from the maximum processing amount setting unit 3 per unit area. In the example of the super-resolution processing of the SAR image using MUSIC, the unit area is a row, and the maximum processing number is the maximum value of the number of scattering points estimated in each row. From the image information reference type area division method setting means 67, the size of the entire image processing area and the allocation method of the processing areas are obtained. The image information reference type area division method setting means 67 includes information on the image processing area (image size, pixel size, data type, etc.) from the processed image information setting means 5, and the number of processors to be used from the number of used processors setting means 4. To get. In the image information reference type area division method setting means 67, the processed image information setting means 5
The image size and the number of processors to be used are obtained from the number-of-processors setting unit 4 to determine the method of allocating the processing areas. This calculation method can be arbitrarily set. In this example, assuming the processing load on each row is uniform,
This is a processing area allocation method based on a general dynamic area partitioning method. However, in this apparatus, a reallocation request is made, and an image area to be processed may increase in the middle. Therefore, in the image information reference type area division method setting means 67,
The allocation method is selected on the assumption that the image area is increased by the reallocation request. Next, the image information reference type image area allocating means 65 initializes the internal information using the obtained information. Here, the "number of remaining processing areas" is calculated from the "size of the entire image processing area" and set to the number of processing units (the number of rows in this example), and the execution flags of all processors are set to "ON". To do. Furthermore, from the "size of the entire image processing area", create an execution management table for each processing unit,
The management parameter for all processing units is set to "not yet", and the head of the "image allocation pointer" is set to the head of the image area (see FIG. 47). Further, it instructs the image information reference type processing load calculation means 66 of all the processors to start the execution (step ST
101). After the initialization up to this point is completed, the image information reference type image area allocating means 65 is in the “request waiting” state from the image information reference type processing load calculating means 66 of each processor (step ST102).

The operation when the "next process" is requested from the image information reference type processing load calculating means 66 in the "request waiting" state will be described. First, the "number of remaining processing areas" is checked (step ST103). If the "number of remaining processing areas" is greater than 0, the next processing area allocation operation is performed. In the processing area allocation operation, first, the “next processing area” to be allocated to the image information reference type processing load calculation means 66 of the request source is determined (step ST104). The size of the "next processing area" to be assigned here follows the method of allocating the processing area designated by the image information reference type area division method setting means 67.
At the time of initial allocation, from the position indicated by the “image allocation pointer” (the beginning of the image area), select the areas for which the management parameter is “not yet” for the “size of the next processing area”, and select the “next processing area”. As ". When the end of the processing area is reached during selection, the area is returned to the beginning and areas having the management parameter of "not yet" are sequentially selected. At the time of reallocation, the image information recorded in the management parameter of the execution management table for each processing unit corresponding to the reallocation area (in the example of the super-resolution processing of the SAR image using MUSIC, the calculation result of the estimated scattering points) Based on, the "subsequent processing area" is determined using the area division method provided by the image information reference type area division method setting means 67. When the "next processing area" is determined, the management parameter of the execution management table for each processing unit corresponding to the area is set to "total" (calculating). Also,
The "image allocation pointer" is advanced to the element for which the management parameter is "not yet". If the end of the processing area is reached,
Return to the beginning and search, and proceed to the first element that is "not yet". If there is no element that is "not yet", the image allocation pointer is set to "empty". Also, "number of remaining processing areas"
Is updated ("number of remaining processing areas"-"size of next processing area") (step ST105). Next, in the processing area allocation operation, the processing restriction amount calculation means 7 is instructed to calculate the “processing restriction amount” for the selected “next processing area” (step ST106). The processing limit amount calculating means 7 obtains the maximum number of processings per unit area from the maximum processing amount setting means 3 per unit area, and from the “next processing area” instructed by the image information reference type image area allocating means 65, The "processing limit amount" in the target area is calculated. In the processing area allocation operation,
After receiving the calculation result of the “processing restriction amount” from the processing restriction amount calculating means 7, the information of the “next processing area” and the “processing restriction amount” is notified to the requesting image information reference type processing load calculating means 66. Yes (step ST107).

The case where the "remaining processing area number" is 0 when the "remaining processing area number" is checked will be described. in this case,
First, the requesting image information reference type processing load calculating means 66 is instructed to "pause" the execution. Further, the in-execution flag of the processor corresponding to the request source image information reference type processing load calculation means 66 is set to "OFF" (step ST1).
08). Next, all of the in-execution flags of the processor are “O”.
It is checked whether it is "FF" (step ST10).
9). Here, if there is even one processor whose running flag is "ON", it returns to the "request waiting" state. On the other hand, when all the execution flags are “OFF”, the image information reference type processing load calculation means 6 of all the processors
6 is notified of "execution completed" (step ST11)
0).

The operation when "reallocation" is requested from the image information reference type processing load calculating means 66 in the "request waiting" state will be described. In this case, first, the “number of remaining processing areas” is updated (“number of remaining processing areas” + “size of reallocation area”) (step ST111). Here, when the image allocation pointer is “empty”, the management parameter is the numerical value of the image information (in the example of the super-resolution processing of the SAR image using MUSIC, the calculation result of the estimated scattering point). An image allocation pointer is set at the beginning of the (step ST112). Next, the in-execution flag of each processor is checked, and if there is a processor that is "OFF", the image information reference type processing load calculation means 6 corresponding to this
6 is notified of "restart execution" (step ST113).
In addition, the in-execution flag corresponding to the processor that has notified the resumption of execution is set to "ON" (step ST114).
After this, the state returns to the "waiting for request" state.

The operation of the image information reference type processing load calculating means 66 will be described with reference to FIG. The image information reference type processing load calculating means 66 first receives an instruction to start execution from the image information reference type image area allocating means 65 (step ST121), and requests the next processing (step ST).
122) and waits for the answer (step ST123). Here, when the answer is the notification of the information of the "next processing area" and the "processing limit amount", a series of processing execution operations are started. In the case of initial allocation (step ST124), in the process execution operation, first, within the processing region designated by the "next processing region", estimation calculation of the scattered point number is executed for each row (step ST12).
5). In the calculation of the estimated number of scattered points, the processing portion 12 that determines the processing load of the image processing program 11 is used.
In this calculation, the maximum number of scattered points per line of the maximum processing amount setting means 3 per unit area is referred to so that the maximum value of the estimated scattered points in each line is equal to or less than this maximum scattered point. Further, the calculation result of the estimated number of scattered points is recorded in the management parameter of the corresponding execution management table for each processing unit. Next, the estimated scattering points in each row are added one by one in order from the first row of the "next processing area", and this sum is compared with the "processing limit amount" (steps ST126 to ST130). Here, a line whose sum exceeds the "processing limit amount" is set as a "boundary line" (step ST131).
From the beginning of the "next processing area" to the "boundary line"("boundaryline"
Area (including) is set as a "super-resolution execution area" (step ST132). After setting the "super-resolution execution region", set the region from the line next to the "boundary line" to the last line of the "next processing region" to the "super-resolution unprocessed region" (see FIG. 48).
(Step ST134). Here, when the "boundary row" is the last row of the "next processing area" (the "reallocation area" is empty), the process proceeds to the next super-resolution processing execution (step ST13).
3). On the other hand, if the “boundary row” is not the last row of the “next processing area” (the “reallocation area” exists), the image information reference type image area allocating means 65 is notified of the “super-resolution unprocessed area”. Request reallocation (step ST135). After notifying this request, the process proceeds to the next execution of super-resolution processing. On the other hand, in the case of reallocation (step ST124), since the estimation calculation of the scattered point number has already been executed, the next super solution is set after the entire area of the “next processing area” is set to the “super resolution execution area”. Proceed to image processing execution (step ST136). In the execution of the super-resolution processing, the image processing execution means 9 is instructed to execute the super-resolution processing for the "super-resolution execution area" (step ST137).
The image processing execution means 9 uses the processing portion 13 of the image processing program 11 whose load changes dynamically to execute the super-resolution processing. After that, after the completion of execution of the super-resolution processing in the image processing execution means 9, the processing execution operation is ended, the process returns to the "next processing request" and the processing is started (step ST
138).

If the reply to the request for the next process is "temporary stop" of execution, the state becomes "waiting for instruction" (step S).
T139). In the “waiting for instruction” state, when the “restart execution” is instructed by the image information reference type image area allocating means 65, the process returns to the “next processing request” and restarts the processing. On the other hand, in the “waiting for instruction” state, when the “execution completion” is instructed from the image information reference type image area allocating means 65, the processing is ended (step ST140).

The operation of the image information reference type area division method setting means 67 will be described. If the cumulative estimated scatter points obtained by accumulating the estimated scatter points calculated in each row of the "next processing area" exceeds the "processing limit amount", the next row of the "boundary row" to the "next processing area" The image information reference type processing load calculation means 66 requests the image information reference type image area allocation means 65 to perform reallocation, with the area up to the last row as the “super-resolution unprocessed area”. Then, the image information reference type image area allocation means 65 uses the area division method given by the image information reference type area division method setting means 67 based on the estimated number of scattering points recorded in the management parameter of the execution management table, Processing area ". The area division method provided by the image information reference type area division method setting means 67 will be described. First,
The "assigned scattered score" is calculated based on the cumulative value of the scattered score recorded in the management parameter of the execution management table and the used processor number set in the used processor number setting means 4. Then, the sum of the estimated scattering points of each row is calculated from the head of the image allocation pointer, and the area up to the row exceeding the “assigned scattering points” is set as the “next processing area”.

A case where the sample image shown in FIG. 9 is processed by a computer having a four-processor configuration in the apparatus according to the thirtieth embodiment will be described. In FIG. 9, the processes are 1, 3,
In the 5th and 7th lines, the processing is concentrated, with the estimated number of scattered points being 100, and in the other rows, the estimated number of scattered points is 1, which is a small process. Here, the calculation time for the estimated number of scattering points is 1, and the time for the super-resolution processing per scattering point is 1. Hereinafter, the unit of the processing time here is “T”. When the processing of FIG. 9 is sequentially executed, “600T”
It will take time. Note that, in FIG. 9, the estimated scatter points and the processing load (processing time) of each line are clearly shown, but this is a value found after the estimation calculation result of the scatter points, and a value that cannot be known before the start of the calculation. Is. Guided
In the example of -Self scheduling (the case of executing without changing the allocation of the first processing area, see FIG. 7),
The total execution time is about "450T" (only 1.3 times faster). On the other hand, in the example shown in the first embodiment, the processing can be completed in a total execution time of only about "177T" (about 3.4).
It has been doubled in speed).

On the other hand, in the device of the thirtieth embodiment, the following processing is performed. As means for performing various settings, the number of used processors of the used processor number setting means 4 is “4”, and the size of the image processing area of the processed image information setting means 5 is “10”.
Set to line "0". In the maximum throughput setting unit 3 per unit area, the maximum estimated number of scattering points per line is set to "1.
00 "is set. The mounting image is as shown in FIG. In the image information reference type image area allocating means 65, the processing proceeds according to the flowchart of FIG. First, initialization is performed based on the data set in each means. The execution flags of the processors 1 to 4 are set to “ON”, the “remaining processing area number” is set to 100, the management parameters of 1 to 100 rows of the execution management table are set to “not yet”, and the “image allocation pointer” is set to 1 row. Set on the eyes.
It is assumed that the image information reference type area division method setting means 67 has notified ““ the size of the entire image processing area ”/“ the number of used processors ”” as a method of determining the “next processing area” size at the time of initial allocation. , Image information reference type processing load calculation means 66
Instruct execution start to -1 to 66-4. The image information reference type processing load calculation means 66-1 to 66-4 proceed with the processing according to the flowchart of FIG. Here, in response to an instruction to start execution from the image information reference type image area allocating means 65, the “next processing request” is sequentially issued to the image information reference type image area allocating means 65. Here, in order to simplify the explanation, first, the image information reference type processing load calculation means 66-1 first makes a “next processing request” to the image information reference type image area allocation means 65, and subsequently at almost the same time. , Image information reference type processing load calculation means 66-2, image information reference type processing load calculation means 66-3, image information reference type processing load calculation means 66
It is assumed that the "next processing request" is made in the order of -4. Further, the image information reference type processing load calculation means 66-1 to 66-
It is assumed that No. 4 simultaneously makes "next processing request" at time "0T". First, at the time “0T”, the image information reference type image area allocation unit 65 causes the image information reference type processing load calculation unit 66-
The "next processing request" from 1 is received. First, according to the area division method of the image information reference type area division method setting means 67,
The first 25 rows and the first 25 rows are selected as the "next processing area". Next, the "remaining processing area number" is updated to 75, the management parameters of the first to 25th lines of the execution management table are updated to "total", and the "image allocation pointer" is updated to the 26th line. After that, the processing limit amount calculation means 7 is caused to calculate the “processing limit amount”. In this example, "the size of the next processing area" *
4-1 "is calculated as the" processing limit amount ".
Since the “size of the next processing area” is 25, the “processing restriction amount” is 99 here. After the calculation of the “processing limit amount”, the “next processing region” (1 to 25 lines) and the “processing limit amount” 99 are notified to the image information reference type processing load calculating means 66-1.
Return to the "waiting for request" state. In the image information reference type image area allocating means 65, the image information reference type processing load calculating means 66-2, the image information reference type processing load calculating means 66-3, and the image information reference type processing load calculating means 6 at time “0T”.
The "next processing request" from 6-4 is received. Here, the "next processing area" is allocated in the same procedure as when the image information reference type processing load calculation means 66-1 requests. The “next processing area” (lines 26 to 50) and the “processing limit amount” 99 are input to the image information reference type processing load calculating means 66-2, and the “next processing area” 99 is input to the image information reference type processing load calculating means 66-3. The “next processing area” (lines 51 to 75) and the “processing restriction amount” 99, and the image information reference type processing load calculating means 66-4 stores the “next processing area” (lines 76 to 100) and the processing restriction. The amount "99" is notified, and the state returns to the "waiting for request" state. As a result of the processing here, the "number of remaining processing areas" is set to 0, and from the 26th row to the 100th row.
The management parameter up to the line becomes "total" and the line pointed by the "image allocation pointer" disappears (there is set again when there is a reallocation request next time).

Image information reference type processing load calculation means 66-1
Then, at time “0T”, the “next processing area” (lines 1 to 25) and the “processing limit amount” 99 are received from the image information reference type image area allocating means 65 as a response to the “next processing request”. First, the estimated scatter points are calculated in each line of the "next processing area", and the calculation result of the estimated scatter points is recorded in the management parameters of lines 1 to 25 of the execution management table. Here, it takes "25T" to calculate the estimated number of scattered points for 25 lines. Next, the sum of the estimated number of scattered points in each row is calculated and compared with the “processing limit amount” 99. Here, the estimated number of scattering points is 100 in the first line, and the first line is the “boundary line”. Therefore, the "super-resolution execution area" is only the first row. Naturally, the "boundary line" is not the last line, so 2 to 2
The reallocation of five lines is requested to the image information reference type image area allocating means 65. Here, at the time “25T” when the calculation of the estimated number of scattered points is finished, the image information reference type image area allocating means 65.
A request for reassignment to After the reallocation request, the image information reference type processing load calculating means 66-1 is operated by the image processing executing means 9
-1 is instructed to execute the super-resolution processing of the "super-resolution execution area" (first row), and the management parameter of the "super-resolution execution area" (first row) is set to "total". In the image information reference type processing load calculation means 66-1, the end of this processing is time “125T”.
Will wait until. The image information reference type processing load calculation means 66-2 receives the “next processing area” (lines 26 to 50) and the “processing limit amount” 99 at time “0T”. The processing proceeds in the same procedure as the image information reference type processing load calculating means 66-1, but here, the sum of the estimated scattering points of each row increases only to 25 even if the calculation is performed up to the last row.
Therefore, the reallocation request is not issued, and all of the 26th to 50th rows which are the "next processing area" are set to the "super-resolution execution area". After setting the “super-resolution execution area”, the image processing execution means 9
-2 is instructed to execute the super-resolution processing of lines 26 to 50. Here, the end of this processing is waited until time "50T". Image information reference type processing load calculation means 66-
3, the "next processing area" (lines 51 to 75) at time "0T"
Line) and the “processing limit amount” 99. Here, as in the case of the image information reference type processing load calculation means 66-2, since the cumulative number of estimated scattering points increases only to a value smaller than 25, which is a “processing limit amount”, lines 51 to 75 are “super-resolution”. Then, the image processing execution means 9-3 is instructed to execute the super-resolution processing. The end of this process will be waited until time “50T”. Image information reference type processing load calculation means 6
In 6-4, at the time “0T”, the “next processing area” (from line 76 to
(100 lines) and "processing limit amount" 99. Here, as in the case of the image information reference type processing load calculation means 66-2, since the cumulative number of estimated scattering points increases only to a value smaller than 25, which is the “processing limit amount”, lines 76 to 100 are “super-resolution”. Image processing execution means 9-
4 is instructed to execute the super-resolution processing. The end of this process will be waited until time “50T”.

At time "25T", the image information reference type processing load calculating means 66-1 makes a reassignment request for 2 to 25 lines (24 lines). Here, the image information reference type image area allocating means 65 updates the “number of remaining processing areas” from 0 to 24 (2
Add 4 lines). Here, since the in-execution flags of the processors 1 to 4 are all “ON”, the process related to the resumption of execution is not performed and the state returns to the “waiting for request” state. Since the execution of the super-resolution processing ends at time “50T”, the image information reference type processing load calculation means 66-2 to image information reference type processing load calculation means 66-4 make a “next processing request”. Here, for simplification of description, it is assumed that the “next processing request” is made in the order of the image information reference type processing load calculation means 66-2 to the image information reference type processing load calculation means 66-4. The image information reference type area division method setting means 67 determines, as a method of determining the “next processing area” at the time of reallocation, ““ assigned scattered point number ”=“ cumulative estimated scattered point number ”+“ number of used processors ”−1 /“ used ” It is assumed that the number of processors "" is notified.
The “cumulative management parameter” of lines 2 to 25, which is the remaining processing area, is 321. Here, the "image allocation pointer" has moved from the beginning of the image area to the line where the numerical value of the estimated scattering point is recorded in the management parameter, that is, the second line. The “assigned scattering score” of the image information reference type processing load calculation means 66-2 is 81. When the sum of the estimated scattering points recorded from the second line is calculated, it becomes 101 in the third line,
It is 81 or more of the "assigned scattered points". Therefore, the 2nd to 3rd rows are set as the "next processing area". The "processing limit amount" is 7, but the estimated scattering point number has already been calculated, so it is not used. The image information reference type processing load calculation means 66-2 which has received the "next processing area" sets the "next processing area" to the "super-resolution processing execution area", and advances the processing to the image processing execution means 9-2.
Instruct execution of the super-resolution processing of 2nd to 3rd rows. here,
The end of this process is waited until "151T". Similarly, the “assigned scattered point number” of the image information reference type processing load calculation means 66-3 is 55. When the sum of the estimated scattered points recorded from the 4th line indicated by the “image allocation pointer” is calculated, it becomes 101 on the 5th line, which is 55 of the “allocated scattered points”.
That is all. Therefore, the 4th to 5th rows are the "next processing area".
And The image information reference type processing load calculation means 66-3 which has received the "next processing area" sets the "next processing area" to the "super-resolution processing execution area", and then to the image processing execution means 9-3.
Instruct execution of super-resolution processing in lines 4 to 5. here,
The end of this process is waited until "151T". Similarly, the “assigned scattered point number” of the image information reference type processing load calculation means 66-4 is 30. When the sum of the estimated scattered points recorded from the 6th line indicated by the “image allocation pointer” is calculated, it becomes 101 on the 7th line, which is 30 of the “allocated scattered points”.
That is all. Therefore, the 6th to 7th lines are the "next processing area".
And The image information reference type processing load calculation means 66-4 which has received the "next processing area" sets the "next processing area" to the "super-resolution processing execution area", and then to the image processing execution means 9-4.
Instruct execution of super-resolution processing in lines 6 to 7. here,
The end of this process is waited until "151T".

Since the super-resolution processing ends at time "125T", the image information reference type processing load calculation means 66-1.
Performs a "next processing request". The “assigned scattering score” of the image information reference type processing load calculation means 66-1 is 5. When the sum of the estimated scattered points recorded from the 8th line indicated by the “image allocation pointer” is calculated, it becomes 5 on the 12th line, which is 5 or more of the “allocated scattered points”. Therefore, 8 rows to 12
The line is the "next processing area". The image information reference type processing load calculation means 66-1 which has received the “next processing area” sets the “next processing area” to the “super-resolution processing execution area”, and sends it to the image processing execution means 9-1 in 8 lines. Instruct execution of super-resolution processing of 12 lines. Here, the end of this process is waited until "130T". Since the execution of the super-resolution processing ends at time “130T”, the image information reference type processing load calculation means 6
6-1 makes a "next processing request". The “assigned scattering score” of the image information reference type processing load calculation means 66-1 is 4.
When the sum of the estimated scattered points recorded from the 13th line indicated by the “image allocation pointer” is calculated, it becomes 4 on the 16th line, which is 4 or more of the “allocated scattered points”. Therefore, 1
Lines 3 to 16 are referred to as "next processing area". "Next processing area"
Image information reference type processing load calculation means 66-1
Sets the "next processing area" to the "super-resolution processing execution area" and instructs the image processing execution means 9-1 to execute the super-resolution processing of lines 13 to 16. Here, the end of this processing is waited until "134T". Similarly, the image information reference type processing load calculation means 66-1 executes the super-resolution processing up to the last line (25 lines) of the “reallocation area”, and at the time “143T”, the “reallocation area” is reached. Becomes empty. The execution of the image information reference type processing load calculation means 66-2 to the image information reference type processing load calculation means 66-4 is also completed at "151T".

Thus, in the example of the image of FIG. 9 this time,
In the device of the thirtieth embodiment, the processing can be completed in only 151T (the speed can be increased by about 4.0 times).

As described above, according to the thirtieth embodiment, the processing areas for reallocation are determined based on the image information obtained from the input image, so that the processors executing in parallel can be efficiently loaded. There is an effect that a parallel image processing device that can be dispersed can be obtained.

Embodiment 31. The image information reference type image area allocating means 65 according to the thirty-first embodiment, when reallocating the processing areas by referring to the image information, prioritizes the number of unit areas to determine the processing areas to be re-allocated. It is configured in. Other configurations are the same as those in the thirtieth embodiment.

Next, the operation will be described. When there is a "reassignment request", the image information reference type image area allocating means 6
5 determines the "next processing area" by using the area division method provided by the image information reference type area division method setting means 67 based on the estimated number of scattering points recorded in the management parameter of the execution management table. At that time, the rows are adopted so that the sum of the estimated scattered points of each row is equal to or more than the “assigned scattered points”, and the area formed by the adopted rows is defined as the “next processing area”. In the thirtieth embodiment, sequential rows are adopted from the beginning of the image allocation pointer. In the thirty-first embodiment, instead of adopting sequentially, starting from the head of the image allocation pointer, the rows to be adopted and the rows not to be adopted are selected based on the estimated scattered points, and the “allocated scattered points” are selected. A row having a value as close as possible to is selected, and this area is set as the "next processing area". The "assigned scattered score" is calculated based on the Guided-Self scheduling, but the efficiency is improved by setting the "next processing area" to be reallocated to each processor as close to the "assigned scattered score" as possible. It can be expected that the load of the remaining processing area is well allocated.

As described above, according to the thirty-first embodiment, when the image information reference type image area allocating means 65 refers to image information to reallocate a processing area, the number of unit areas is prioritized. By deciding the processing area to be reallocated, the parallel image processing apparatus capable of efficiently allocating the load of the remaining processing area can be obtained.

Embodiment 32. The image information reference type image area allocating means 65 according to the thirty-second embodiment determines the processing area to be re-allocated by prioritizing the size of the image information when re-allocating the processing area by referring to the image information. It is configured to do. Other configurations are the same as those in the thirtieth embodiment.

Next, the operation will be described. When there is a "reassignment request", the image information reference type image area allocating means 6
5 determines the "next processing area" by using the area division method provided by the image information reference type area division method setting means 67 based on the estimated number of scattering points recorded in the management parameter of the execution management table. At that time, the rows are adopted so that the sum of the estimated scattered points of each row is equal to or more than the “assigned scattered points”, and the area formed by the adopted rows is defined as the “next processing area”. In the thirtieth embodiment, sequential rows are adopted from the beginning of the image allocation pointer. In the thirty-second embodiment, instead of sequentially adopting, the one having the largest estimated scattering point number is sequentially adopted. As a result, even when there is a row having a large estimated scattering point at a position far from the beginning of the image allocation pointer in the remaining processing area, the load of the remaining processing area can be expected to be allocated efficiently.

As described above, according to the thirty-second embodiment, when the image information reference type image area allocating means 65 refers to the image information to reallocate the processing area, the size of the image information is changed. By deciding the processing area to be reallocated preferentially, there is an effect that a parallel image processing apparatus capable of efficiently allocating the load of the remaining processing area can be obtained.

Embodiment 33. The image information reference type image area allocating means 65 according to the thirty-third embodiment does not simply adopt a processing area for performing reallocation from image information when the processing area is reallocated by referring to the image information. The processing load of the image processing to be executed is calculated based on the obtained image information, and the processing area to be reallocated is determined based on the processing load. Other configurations are the same as those in the thirtieth embodiment.

Next, the operation will be described. When there is a “reassignment request”, in the above-mentioned Embodiments 30 to 32,
The image information reference type image area allocation means 65 uses the area division method given by the image information reference type area division method setting means 67 based on the estimated scattering points recorded in the management parameters of the execution management table to indicate “the next processing area”. Is determined. this is,
This is because it is known that the execution time of the super-resolution processing depends on the estimated number of scattering points. However, after performing the "calculation of the estimated number of scattering points → super-resolution processing" once, the relationship between the number of estimated scattering points and the execution time of the super-resolution processing has been clarified. Therefore, the "next processing region" at the time of reallocation is determined based on a value obtained by converting the estimated scattering point into the execution time of the super-resolution processing, not simply the estimated scattering point. As a result, it can be expected that the load of the remaining processing area can be efficiently allocated.

As described above, according to the thirty-third embodiment, when the image information reference type image area allocating means 65 refers to the image information and reallocates the processing area, the image information is simply re-allocated from the image information. Rather than adopting the processing area to allocate, by calculating the processing load of the image processing to be executed based on the obtained image information, by determining the processing area to be reallocated based on the processing load, There is an effect that it is possible to obtain a parallel image processing apparatus that can efficiently allocate the load of the remaining processing area.

Embodiment 34. The image information reference type processing load calculation means 66 according to the thirty-fourth embodiment is configured to continuously execute the image information calculation and the processing instruction to the image processing execution means 9 at the time of initial allocation. is there.
Other configurations are the same as those in the thirtieth embodiment.

Next, the operation will be described. In the thirty-third embodiments described above, at the time of initial allocation, for each row of the allocated "next processing area", first, all the estimated scattering point numbers are calculated, and then the super-resolution processing is executed.
In the thirty-fourth embodiment, regarding the initial allocation, the “calculation of estimated scattered points → super-resolution processing” is continuously executed for each allocated row. This allows
It can be expected that the cache information will be used efficiently and the execution time will be reduced.

As described above, according to the thirty-fourth embodiment, the image information reference type processing load calculating means 66 continuously calculates the image information and instructs the image processing executing means at the time of initial allocation. By executing the above, there is an effect that a parallel image processing apparatus capable of efficiently using cache information and reducing execution time can be obtained.

[0258]

As described above, according to the present invention, the area division method setting means for setting the processing area allocation method according to the number of used processors and the image processing area size, and the maximum processing load amount in the unit area And a processing limit amount calculating unit that sets a processing limit amount according to the calculated processing region and a processing region based on the allocation method set by the region division method setting unit, and the processing region and the processing region are calculated. An image in which the processing restriction amount set according to the processing restriction amount calculation means is allocated to each processor, and when the processor requests re-allocation, the area requested for re-allocation is re-allocated as an unprocessed area. In the area allocation means and each processor, the processing load in the target image area is calculated with respect to the processing area allocated by the image area allocation means. Instructing the corresponding image processing executing means to perform processing in a range that does not exceed the specified processing limit amount, and when the processing limit amount is exceeded, request the image area allocating means to reallocate the over-limit area. Since the processing load calculation means is provided, the processing load assigned to each processor is monitored on the basis of the processing limit amount according to the maximum processing load amount in the unit area, and the process at a certain processor is processed. If the number exceeds the processing limit, the processing area assigned to the processor that has exceeded the processing limit amount is reassigned, so that the algorithm for image processing in which the processing load changes depending on the input image can be efficiently executed.

According to the present invention, the area division method setting means sets the processing area allocation method according to the maximum processing load amount in the unit area set in the maximum processing amount setting means per unit area. Since it is configured as the type area division method setting means, there is an effect that a processing area of an appropriate size can be assigned to each processor according to the maximum processing load amount in the unit area.

According to the present invention, the area division method setting means is provided with processing load measuring means for actually measuring the execution time for the maximum load and the execution time for the minimum load for the unit area in the image processing algorithm, and the area division method setting means is set. The means is a measurement corresponding area division method setting means for setting the allocation method of the processing area according to the actual measurement time measured by the processing load measuring means, and the processing limit amount calculating means is the actual measurement measured by the processing load measuring means. Since it is configured to be a measurement-compatible process limit amount calculation unit that sets a process limit amount according to time, a processing region of an appropriate size is set based on the maximum and minimum execution times in the measured unit region. There is an effect that a processing limit amount can be assigned to each processor.

According to the present invention, the processing block setting means for setting an image area of an arbitrary size as a block is provided, and the processing load calculation means is adapted to the processing block corresponding to the block set by the processing block setting means. Since the image processing execution means is the processing load calculation means and the image processing execution means is the processing block correspondence image processing execution means for executing the image processing of the image region designated by the processing block correspondence processing load calculation means, The execution efficiency can be improved by preferentially processing the image area in the size block.

According to the present invention, according to the cache information setting means for setting the cache information in the target computer, the image processing area size set by the processing image information setting means and the cache information set by the cache information setting means. Since the processing block setting means for setting the block size set by the processing block setting means is provided, it is possible to set the block size according to the image processing area size and the cache information. .

According to the present invention, the size of the block set by the processing block setting unit according to the processing region calculated by the image region allocating unit and the size of the block designated by the processing block corresponding processing load calculating unit. Since it is configured to include a dynamic processing block calculation means for dynamically changing and setting the block size, there is an effect that the block size can be set more appropriately.

According to the present invention, the area division method setting means is the processing block corresponding area division method setting means for setting the processing area allocation method according to the size of the block set by the processing block setting means. Since it is configured as described above, there is an effect that the calculation of the range specified by the block can be executed more appropriately.

According to the present invention, the processing limit amount calculating means is configured as the processing block corresponding processing limit amount calculating means for setting the processing limit amount according to the size of the block set by the processing block setting means. Therefore, there is an effect that the calculation of the range specified by the block can be executed more appropriately.

According to the present invention, the area division setting means is
A unit area division correspondence type area division setting means for setting a processing area allocation method in consideration of division of a unit area, and a processing limit amount calculation means for setting a processing limitation amount when a unit area is divided. As a type processing limit calculation means,
The image area allocating means is a unit area division correspondence type image area allocating means for allocating and reallocating the processing areas by dividing the unit areas, and the processing load calculating means calculates the processing load corresponding to the division of the unit areas. And a unit area division-compatible processing load calculating means for determining presence / absence of reallocation, and an image processing executing means executes image processing of an image area designated corresponding to division of a unit area. Since the processing execution means is configured, when the processing load on a certain processor increases, the unit area can be divided and the processing can be reallocated to another processor, and the processing load changes depending on the input image. Even when the processing load in a certain unit area is significantly increased by the algorithm for image processing, there is an effect that the load can be efficiently distributed to the processors executing in parallel.

According to the present invention, the unit area division corresponding to the unit area division corresponding processing load calculation means is determined by referring to the information of the unit area division correspondence type area division setting means, and the possibility of division of the unit area is determined. Since it is configured to include the permission / prohibition condition setting means, it is possible to select the load distribution suitable for the current execution status, and it is possible to improve the execution efficiency.

According to the present invention, when a unit area is divided and reallocated, the calculation result of the unit area division-compatible processing load calculating means that first calculates the processing load in the unit area is retained and divided. The processing load calculation time is reduced because the processing load calculation means for reallocating the allocated unit area to the unit area division-compatible processing load calculation means of another processor can refer to the calculation result. There is an effect that it can be saved and the execution efficiency can be improved.

According to the present invention, according to the cache information setting means for setting the cache information in the target computer, the cache information set in the cache information setting means and the image processing area size set in the processed image information setting means. And a unit area division cost calculation means for calculating a cost when the unit area is divided and calculating a processing limit amount from the calculated cost and transmitting it to the unit area division correspondence processing load calculation means. Since it is configured, there is an effect that the calculation in the area after the division can be executed in an efficient range and the execution efficiency can be improved.

According to the present invention, the load prediction setting means for setting the prediction load for each unit area in the image processing algorithm is provided, and the area division method means is used in accordance with the prediction load set by the load prediction setting means. Since the load prediction type area dividing method means for setting the processing area allocation method is configured, the processing load can be more evenly distributed and the execution efficiency can be improved.

According to the present invention, the brightness value calculation formula setting means for setting the calculation formula of the brightness value of the target image, and the brightness for each pixel of the target image according to the calculation formula set by the brightness value calculation formula setting means. A luminance value calculation executing means for calculating a value, and a pixel reference load prediction calculating means for calculating a predicted load according to the brightness value calculated by the luminance value calculation executing means and transmitting it to the load prediction setting means are configured. Therefore, when the accuracy of the prediction load from the brightness value of each pixel is high, there is an effect that the prediction accuracy of the load can be increased.

According to the present invention, the pixel reference load prediction condition setting means for setting the calculation method of the prediction load for the brightness value of each pixel is provided, and the pixel reference load prediction calculation means is set to the pixel reference load prediction condition setting means. Since it is configured to be the measurement condition corresponding pixel reference load prediction calculation means for calculating the prediction load for each unit area according to the calculated calculation method, the calculation method of the prediction load can be flexibly set from the brightness value of each pixel. There is an effect that can be.

According to the present invention, the characteristic region extraction setting means for setting the calculation method of the characteristic region serving as the load prediction reference from the luminance value of the pixel, and the luminance according to the calculation method set by the characteristic region extraction setting means. A characteristic region extracting unit that extracts a characteristic region from the brightness value of each pixel of the target image calculated by the value calculation executing unit, and the pixel reference load prediction calculating unit is configured to operate in accordance with the characteristic region extracted by the characteristic region extracting unit. The image processing algorithm is configured to calculate the predicted load for each unit area and transmit it to the load prediction setting unit as the characteristic region load prediction calculation unit. There is an effect that can increase.

According to the present invention, the characteristic region load prediction condition setting means for setting the calculation method of the predicted load for the characteristic region is provided, and the characteristic region load prediction calculation means is set to the calculation set by the characteristic region load prediction condition setting means. Since it is configured as the measurement condition corresponding characteristic region load prediction calculation unit that calculates the predicted load for each unit area according to the method, there is an effect that the calculation method of the predicted load can be flexibly set.

According to the present invention, the reference area setting means for setting an image area of a certain range as the reference area is provided, and the pixel reference load prediction calculation means is used to calculate the brightness value in the reference area set by the reference area setting means. Since the reference region load prediction calculation unit calculates the prediction load for each unit region in the image processing algorithm according to the brightness value calculated by the execution unit, the prediction load is calculated from the brightness value of the image region in a certain range. By doing so, there is an effect that the prediction accuracy of the load can be improved.

According to the present invention, the reference area load prediction condition setting means for setting the calculation method of the predicted load according to the brightness value in the reference area is provided, and the reference area load prediction calculation means sets the reference area load prediction condition setting. It is configured so that the reference area load prediction calculation means corresponding to the measurement condition calculates the predicted load for each unit area according to the calculation method set by the means, and thus the predicted load is calculated from the brightness value of the image area within a certain range. By selecting the method, there is an effect that the load prediction accuracy can be improved.

According to the present invention, the inclination reference setting means for setting the inclination of the reference luminance value and the pixel reference load prediction calculation means are provided with the unit area according to the inclination of the luminance value set by the inclination reference setting means. Since the gradient load prediction calculation means for calculating the predicted load for each is configured, the load prediction accuracy can be improved by predicting the load based on the brightness value gradient.

According to the present invention, the slope load prediction condition setting means for setting the calculation method of the prediction load for the slope of the luminance value is provided, and the slope load prediction calculation means is set to the calculation method set by the slope load prediction condition setting means. Since it is configured to be the measurement condition-compatible gradient load prediction calculation means that calculates the predicted load for each unit area according to the above, it is possible to improve the load prediction accuracy by selecting the predicted load calculation method. There is.

According to the present invention, the photographing area load prediction calculation means for calculating the predicted load for each unit area in the image processing algorithm from the area information for photographing the image and transmitting it to the load prediction setting means is provided. Therefore, by predicting the load from the area information of the captured image, it is possible to increase the load prediction accuracy.

According to the present invention, the brightness value calculation formula setting means, the brightness value calculation executing means, the pixel reference load prediction condition setting means and the measurement condition corresponding pixel reference load prediction calculation means, the characteristic region extraction setting means and the characteristics. Area extraction means, characteristic area load prediction condition setting means and measurement condition corresponding characteristic area load prediction calculation means, reference area setting means, reference area load prediction condition setting means, and measurement condition corresponding reference area load prediction calculation means A tilt reference setting means, a tilt load prediction condition setting means and a measurement condition compatible tilt load prediction calculation means, a shooting area load prediction calculation means, and a prediction calculation weight setting means capable of weighting the prediction conditions. Since it is configured as described above, there is an effect that various prediction means can be combined and used flexibly, and the load prediction accuracy can be improved.

According to the present invention, the predicted value recording means for recording the calculated predicted value and the input condition, the processing result recording means for recording the processing load calculated by the actual execution for the predicted value, and the predicted value Since the prediction parameter recorded by the recording unit and the processing load recorded by the processing result recording unit are configured to include a prediction parameter correction unit that corrects the calculation method of the prediction load, the past prediction value and the actual processing load are included. By correcting and improving the calculation method of the predicted load from, there is an effect that the prediction accuracy of the load can be increased.

According to the present invention, the condition for picking up the image at the time of executing the past prediction calculation is recorded, and the optimum correction for the prediction calculation is selected from the image pickup condition for the image for which the new prediction calculation is performed. Since the imaging condition load prediction correction means is provided, the prediction calculation method of the processing load can be corrected when a new prediction is made from the conditions at the time of the past shooting, and the prediction accuracy of the load can be improved. There is an effect that can be.

According to the present invention, when a new predicted value calculation is instructed, a similar condition is searched and selected among the predicted value calculation input conditions recorded in the predicted value recording means, and the processing result is obtained. Since the optimum prediction condition calculation means for correcting the selected prediction calculation with reference to the calculation result recorded in the recording means is provided, the prediction calculation can be corrected and the load prediction accuracy is improved. There is an effect that can be.

According to the present invention, the pre-calculation selection setting means for setting the condition as to whether or not the processing by the pre-calculation is performed,
Pre-calculation determination means for determining the presence or absence of processing by pre-calculation from the conditions set by the pre-calculation selection setting means
A pre-processing load amount calculating means for executing a calculation for obtaining a processing load amount for each unit area for the entire image area in advance when the pre-calculation determining means instructs the pre-calculating processing, and the pre-processing load amount calculating means. Based on the processing load amount for each unit area calculated in advance by the above, since it is configured to include a pre-processing compatible image area allocation means for allocating areas so that the processing load of each processor becomes equal,
There is an effect that it is possible to select and execute the processing method based on the pre-calculation.

According to the present invention, the area division setting means is
An image information reference type area division setting means for setting a method of allocating a processing area in consideration of image information, and the image area allocating means refers to the image information to reallocate the processing area when the processor requests the reallocation. The image information reference type image area allocating means for allocating, and the processing load calculating means is configured to be the image information reference type processing load calculating means for calculating the image information for each processing unit with respect to the allocated processing area. Therefore, by determining the processing area for reallocation based on the image information obtained from the input image, it is possible to efficiently distribute the load to the processors that execute in parallel.

According to the present invention, in the image information reference type image area allocating means, when the processing area is re-allocated by referring to the image information, the unit area number is prioritized to determine the processing area to be re-allocated. Since it is configured to do so, there is an effect that the load of the remaining processing area can be efficiently allocated.

According to the present invention, in the image information reference type image area allocating means, when the processing area is re-allocated by referring to the image information, the processing area is re-allocated by giving priority to the size of the image information. Since it is configured to determine, the load of the remaining processing area can be efficiently allocated.

According to the present invention, in the image information reference type image area allocating means, when the processing area is reallocated by referring to the image information, the processing area to be reallocated from the image information is simply adopted. Instead, the processing load of image processing to be executed is calculated based on the obtained image information, and the processing area to be reallocated is determined based on the processing load. The effect is that the load can be assigned.

According to the present invention, the image information reference type processing load calculating means is configured to continuously execute the image information calculation and the processing instruction to the image processing executing means at the time of initial allocation. The cache information can be used efficiently and the execution time can be shortened.

According to the present invention, when the parallel computer executes the image processing algorithm in which the processing load changes depending on the input image, the allocation processing area is dynamically determined, and
The processing restriction amount for the allocated processing area is calculated from the maximum processing load amount in the unit area, and is allocated to the algorithm execution means of each processor as a set of the allocation processing area and the processing restriction amount, and processed from the algorithm execution means of each processor. An area that exceeds the limit amount is received, the area is incorporated into the entire unprocessed area, and again assigned to the algorithm execution means of each processor as a processing area. In each processor, the target image is assigned to the assigned image area. Calculate the processing load in the area, instruct execution of processing in the range that does not exceed the specified processing limit, and
When the processing limit is exceeded, it is configured to request reallocation of the over-limit area, so the processing load assigned to each processor based on the processing limit according to the maximum processing load in the unit area If the processing by a certain processor exceeds the processing limit amount, the processing area assigned to the processor that has exceeded the processing limit amount is reallocated, and the processing load changes depending on the input image. The effect is that the algorithm can be executed efficiently.

According to the present invention, the allocation method of the allocation processing area is determined from the maximum processing load quantity in the unit area in the image processing algorithm. Therefore, the maximum processing load quantity in the unit area is used as a reference. It is possible to determine a method for calculating the size of the processing area to be allocated to each processor, and thus it is possible to allocate an appropriately sized processing area to each processor according to the maximum processing load amount in the unit area. effective.

According to the present invention, the execution time with respect to the maximum load and the execution time with respect to the minimum load in the unit area of the image processing algorithm are actually measured, and the allocation method of the allocation processing area and the calculation method of the processing limit amount are calculated from the measured time. Since it is configured to determine, the allocation processing area and the processing restriction amount of an appropriate size can be allocated to each processor on the basis of the maximum and minimum execution times in the actually measured unit area.

According to the present invention, an image area of an arbitrary size is designated as a block, and the designated processing is preferentially performed within that block. Therefore, an image area of a block of an arbitrary size is designated. The priority processing has the effect of improving the execution efficiency.

According to the present invention, when the processing load on a certain processor significantly increases and the reallocation based on the unit area in the image processing algorithm is not sufficient, the unit area is divided and the reallocation is performed. Also, each processor is configured to judge the specified processing limit amount and execute the processing in the divided unit area in consideration of the division of the unit area, so that the processing load on a certain processor is increased. When the processing is performed, the unit area can also be divided and the processing can be reassigned to other processors, and the processing load changes depending on the input image. However, there is an effect that the load can be efficiently distributed to the processors that execute in parallel.

According to the present invention, whether or not the unit area can be divided is determined based on the number of waiting processors and the amount of the remaining processing area. Therefore, load distribution suitable for the current execution status can be achieved. There is an effect that it can be selected and the execution efficiency can be improved.

According to the present invention, when the unit area is divided and the reallocation is requested, the processor which processes the unit area first records the calculation result of the first half portion for carrying out the processing load. The remaining processing in the divided unit area is configured so that the processor instructed later uses the recorded calculation result of the first half to perform the latter half of the processing. There is an effect that the calculation time of can be saved and the execution efficiency can be improved.

According to the present invention, when the unit area is divided, the division position where the cache access is advantageous is obtained from the sizes of the cache and the image area. Therefore, the division is performed in an efficient range. There is an effect that the calculation can be executed in the latter area and the execution efficiency can be improved.

According to the present invention, the prediction load is set for each unit area in the image processing algorithm, and the image area allocation method is determined based on the prediction load. Therefore, the processing load is more evenly distributed. There is an effect that it can be distributed and the execution efficiency can be improved.

According to the present invention, the prediction load is calculated on the basis of the brightness value of each pixel of the image to be processed. Therefore, when the accuracy of the prediction load from the brightness value of each pixel is high, the load of the load There is an effect that the prediction accuracy can be improved.

According to the present invention, the characteristic region is determined from the luminance value of each pixel of the image to be processed, and the prediction load is calculated based on the characteristic region. Therefore, the load is predicted by the characteristic region. There is an effect that the load prediction accuracy can be increased.

According to the present invention, the image area within a certain range is set as the reference area, and the prediction load is calculated by using the luminance value of each pixel in the reference area.
There is an effect that the prediction accuracy of the load can be improved by applying the prediction load from the brightness value of the image area in a certain range.

According to the present invention, the calculated predicted value and the input condition for the predicted value are recorded, the processing load calculated in the actual execution is recorded for the predicted value, and the predicted load is based on these recorded values. Since the calculation method is corrected, the calculation method of the predicted load is corrected and improved from the past predicted value and the actual processing load, so that the prediction accuracy of the load can be improved.

According to the present invention, a method for newly calculating a predicted value is selected from the conditions for capturing an image at the time of executing the past prediction calculation and the input condition which is the reference for calculating the predicted value. Since it is configured to perform an appropriate correction, when a new prediction is made from the conditions at the time of shooting in the past, the prediction calculation method of the processing load can be corrected, and the load prediction accuracy can be improved. There is.

According to the present invention, it is determined whether or not the processing load for each unit area is obtained in advance, and when the processing load for each unit area is obtained in advance, the obtained load is used as a reference. Since the method for allocating the image area is determined, there is an effect that it is possible to select and execute the processing method by the pre-calculation.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a parallel image processing device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a process of an image area allocating means.

FIG. 3 is a flowchart showing processing of processing load calculation means.

FIG. 4 is a configuration diagram showing an implementation image of a parallel image processing device during parallel execution.

FIG. 5 is an explanatory diagram showing an outline of an operation of a program that executes super-resolution processing of a SAR image using MUSIC.

FIG. 6 is an explanatory diagram showing a concept of operation of an execution management table.

FIG. 7 is an explanatory diagram showing an example of allocation of processing areas in Guided-Self scheduling.

FIG. 8 is an explanatory diagram showing an example of a “super-resolution execution area” and an “unprocessed area” for a “next processing area” assigned to the processing load calculation means.

FIG. 9 is an explanatory diagram showing an estimated number of scattering points and a processing load (processing time) for a certain sample image in the super-resolution processing of the SAR image using MUSIC which is a processing target.

FIG. 10 is an explanatory diagram showing an implementation image when processing is performed with a four-processor configuration.

FIG. 11 is a configuration diagram showing a parallel image processing device according to a second embodiment of the present invention.

FIG. 12 is a configuration diagram showing a parallel image processing device according to a third embodiment of the present invention.

FIG. 13 is a configuration diagram showing a parallel image processing device according to a fourth embodiment of the present invention.

FIG. 14 is a configuration diagram showing a parallel image processing device according to a fifth embodiment of the present invention.

FIG. 15 is a configuration diagram showing a parallel image processing device according to a sixth embodiment of the present invention.

FIG. 16 is a configuration diagram showing a parallel image processing device according to a seventh embodiment of the present invention.

FIG. 17 is a configuration diagram showing a parallel image processing device according to an eighth embodiment of the present invention.

FIG. 18 is a configuration diagram showing a parallel image processing device according to a ninth embodiment of the present invention.

FIG. 19 is a configuration diagram showing a parallel image processing device according to a tenth embodiment of the present invention.

FIG. 20 is a flowchart showing a process of a unit region division correspondence type image region allocation means.

FIG. 21 is a flowchart showing the processing of a unit area division-compatible processing load calculation means.

FIG. 22 is an explanatory diagram showing an image of division of processing in units of rows.

FIG. 23 is a configuration diagram showing a parallel image processing device according to an eleventh embodiment of the present invention.

FIG. 24 is a configuration diagram showing a parallel image processing device according to a twelfth embodiment of the present invention.

FIG. 25 is a configuration diagram showing a parallel image processing device according to a thirteenth embodiment of the present invention.

FIG. 26 is a configuration diagram showing a parallel image processing device according to a fourteenth embodiment of the present invention.

FIG. 27 is a configuration diagram showing a parallel image processing device according to a fifteenth embodiment of the present invention.

FIG. 28 is a configuration diagram showing a parallel image processing device according to a sixteenth embodiment of the present invention.

FIG. 29 is a configuration diagram showing a parallel image processing device according to a seventeenth embodiment of the present invention.

FIG. 30 is a configuration diagram showing a parallel image processing device according to an eighteenth embodiment of the present invention.

FIG. 31 is a configuration diagram showing a parallel image processing device according to a nineteenth embodiment of the present invention.

FIG. 32 is a configuration diagram showing a parallel image processing device according to a twentieth embodiment of the present invention.

FIG. 33 is an explanatory diagram showing a reference area.

FIG. 34 is a configuration diagram showing a parallel image processing device according to a twenty-first embodiment of the present invention.

FIG. 35 is a configuration diagram showing a parallel image processing device according to a twenty-second embodiment of the present invention.

FIG. 36 is a configuration diagram showing a parallel image processing device according to a twenty-third embodiment of the present invention.

FIG. 37 is a configuration diagram showing a parallel image processing device according to a twenty-fourth embodiment of the present invention.

FIG. 38 is a configuration diagram showing a parallel image processing device according to a twenty-fifth embodiment of the present invention.

FIG. 39 is a configuration diagram showing a parallel image processing device according to a twenty-sixth embodiment of the present invention.

FIG. 40 is a configuration diagram showing a parallel image processing device according to a twenty-seventh embodiment of the present invention.

FIG. 41 is a configuration diagram showing a parallel image processing device according to a twenty-eighth embodiment of the present invention.

FIG. 42 is a configuration diagram showing a parallel image processing device according to a twenty-ninth embodiment of the present invention.

FIG. 43 is a configuration diagram showing a parallel image processing device according to a thirtieth embodiment of the present invention.

FIG. 44 is a flowchart showing the processing of the image information reference type image area allocating means.

FIG. 45 is a flowchart showing the processing of the image information reference type processing load calculation means.

FIG. 46 is a configuration diagram showing an implementation image of a parallel image processing device at the time of parallel execution.

FIG. 47 is an explanatory diagram showing the concept of operation of the execution management table.

[Fig. 48] Fig. 48 is an explanatory diagram showing an example of "super-resolution execution region" and "super-resolution unprocessed region" for the "next processing region" assigned to the image information reference type processing load calculation means.

FIG. 49 is an explanatory diagram showing an implementation image when processing is performed with a four-processor configuration.

FIG. 50 is an explanatory diagram showing the concept of an allocation method when the image information reference type area division method setting means performs reallocation.

[Explanation of symbols]

1 parallel image processing device, 2 image area allocating means, 3 maximum processing amount setting means per unit area, 4 number of processors used setting means, 5 processed image information setting means, 6 area division method setting means, 7 processing limit amount calculating means , 8 processing load calculation means, 9 image processing execution means, 10 platform,
11 image processing program, 12 processing part for determining processing load, 13 processing part whose load changes dynamically, 14
Load calculation type area division method setting means, 15 processing load measuring means, 16 measurement corresponding area dividing method setting means, 17 measurement corresponding processing limit amount calculating means, 18 processing block setting means, 19 processing block corresponding processing load calculating means , 20
Image processing execution means for processing block, 21 cache information setting means, 22 processing block calculating means, 23
Dynamic processing block calculating means, 24 processing block corresponding area dividing method setting means, 25 processing block corresponding processing limit amount calculating means, 26 unit area dividing corresponding image area allocating means, 27 unit area dividing corresponding area dividing setting means ,
28 unit area division corresponding processing limit amount calculation means, 29 unit area division correspondence processing load calculation means, 30 unit area division correspondence image processing execution means, 31 unit area division permission / inhibition condition setting means, 32 pre-calculation result holding means, 33 unit area division cost calculation means, 34 load prediction setting means, 3
5 load prediction type area division method means, 36 brightness value calculation formula setting means, 37 brightness value calculation executing means, 38 pixel reference load prediction calculation means, 39 pixel reference load prediction condition setting means, 40 measurement condition compatible pixel reference load prediction Calculation means,
41 characteristic region extraction setting means, 42 characteristic region extraction means, 4
3 feature area load prediction calculation means, 44 feature area load prediction condition setting means, 45 measurement condition compatible feature area load prediction calculation means, 46 reference area setting means, 47 reference area load prediction calculation means, 48 reference area load prediction condition setting Means, 4
9 measurement condition compliant reference area load prediction calculation means, 50
Inclination reference setting means, 51 Inclination load prediction calculation means, 52
Tilt load prediction condition setting means, 53 Measurement condition compatible tilt load prediction calculation means, 54 shooting area load prediction calculation means, 55 prediction calculation weighting setting means, 56 predicted value recording means, 57 processing result recording means, 58 prediction parameter correction means , 59 shooting condition load prediction correction means, 60 optimum prediction condition calculation means, 61 pre-calculation selection setting means, 6
2 pre-calculation determination means, 63 pre-processing load amount calculation means, 64 pre-processing compatible image area allocation means, 65 image information reference type image area allocation means, 66 image information reference type processing load calculation means, 67 image information reference type areas Division method setting means.

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06T 1/20 G06T 1/20 BF term (reference) 5B005 JJ11 LL15 MM01 UU44 5B045 AA01 GG02 GG14 5B057 CA08 CA12 CA16 CB08 CB12 CB16 CC02 CH02 CH04 CH11 CH18

Claims (47)

[Claims] 1. A parallel image processing apparatus for executing an image processing algorithm in which a processing load changes depending on an input image in a parallel computer, and a maximum processing amount setting unit for setting a maximum processing load amount in a unit area, , The number of used processors setting means for setting the number of used processors, the processed image information setting means for setting the image processing area size, the number of used processors set in the used processor number setting means and the set processed image information setting means Area division method setting means for setting a processing area allocation method according to the image processing area size, and maximum processing load amount in the unit area set in the maximum processing amount per unit area setting means and calculated. A process limit amount calculating means for setting a process limit amount according to a processing area, and an allocation method set by the area dividing method setting means When a processing area is calculated based on the method and the processing area and the processing restriction amount set by the processing restriction amount calculation means according to the processing area are assigned to each processor, and the processor requests re-allocation, , An image area allocating unit that reallocates the area requested to be reallocated as an unprocessed area, and a processing load on the target image area with respect to the processing area allocated by the image area allocating unit in each processor. And instruct processing to the corresponding image processing execution means in a range that does not exceed the specified processing limit amount, and if the processing limit amount is exceeded, reallocate the over-limit area to the image area. A parallel image processing apparatus comprising: a processing load calculation unit that requests an allocation unit. 2. A load calculation type area division method, wherein the area division method setting means sets the allocation method of the processing area according to the maximum processing load amount in the unit area set in the maximum processing amount per unit area setting means. Claim as a setting means
The parallel image processing device described in 1. 3. A processing load measuring means for measuring an execution time for a maximum load and an execution time for a minimum load in the image processing algorithm, the area division method setting means being measured by the processing load measuring means. The measurement-restricted area division method setting means for setting the processing area allocation method according to the measured time, and the processing limit amount calculation means,
2. The parallel image processing apparatus according to claim 1, wherein the parallel image processing device is a measurement corresponding processing limit amount calculating unit that sets a processing limit amount according to an actual measurement time measured by the processing load measuring unit. 4. A processing block corresponding processing load calculation comprising processing block setting means for setting an image area of an arbitrary size as a block, wherein the processing load calculation means preferentially processes the block set by the processing block setting means. The image processing execution means is a processing block correspondence image processing execution means for executing image processing of an image area designated by the processing block correspondence processing load calculation means. Item 5. The parallel image processing device according to any one of items 3. 5. A cache information setting unit for setting cache information in the target computer, a processing block according to the image processing area size set by the processing image information setting unit and the cache information set by the cache information setting unit. 5. The parallel image processing apparatus according to claim 4, further comprising processing block calculation means for setting the size of the block set by the setting means. 6. The size of the block set by the processing block setting means is dynamically set according to the processing area calculated by the image area allocating means and the size of the block specified by the processing block corresponding processing load calculating means. The parallel image processing apparatus according to claim 4 or 5, further comprising a dynamic processing block calculation means for changing and setting. 7. The area division method setting means is a processing block corresponding area division method setting means for setting a processing area allocation method according to a block size set by the processing block setting means. The parallel image processing device according to any one of claims 4 to 6. 8. The processing limit amount calculating means is a processing block corresponding processing limit amount calculating means for setting the processing limit amount according to the size of the block set by the processing block setting means. The parallel image processing device according to any one of claims 4 to 7. 9. A case where the area division setting means is a unit area division correspondence area division setting means for setting a processing area allocation method in consideration of division of a unit area, and the processing limit amount calculating means divides the unit area A unit area division-compatible processing limit amount calculating means for setting the processing restriction amount, and an image area allocating means as a unit area division-compatible image area allocating means for allocating and reallocating processing areas by dividing the unit area. , The processing load calculation means is a processing load calculation means for unit area division corresponding to the division of the unit area to determine the processing load calculation and the presence / absence of reallocation, and the image processing execution means corresponds to the division of the unit area. 2. The parallel image processing apparatus according to claim 1, wherein the parallel image processing apparatus is unit area division correspondence type image processing executing means for executing image processing of a designated image area. 10. A unit area division permission / inhibition condition setting means for determining whether or not a unit area can be divided by referring to the information of the unit area division correspondence type area division setting means and transmitting it to the unit area division correspondence type processing load calculation means. The parallel image processing device according to claim 9, further comprising: 11. When dividing and reallocating a unit area,
First, the processing load is calculated in the unit area, and the calculation result of the unit area division compatible processing load calculation means is held, and the unit area division compatible processing load calculation of another processor to which the divided unit area is reallocated The parallel image processing device according to claim 9 or 10, characterized in that the means comprises a pre-calculation result holding means for making it possible to refer to the calculation result. 12. A unit area according to cache information setting means for setting cache information in a target computer, cache information set in the cache information setting means and image processing area size set in the processed image information setting means. And a unit area division cost calculation means for calculating a cost when the division is performed and calculating a processing limit amount from the calculated cost and transmitting it to the unit area division correspondence processing load calculation means. The parallel image processing device according to any one of claims 9 to 11. 13. A load prediction setting means for setting a predicted load for each unit area in the image processing algorithm, and area dividing method means for setting a predicted load of the processing area according to the predicted load set by the load prediction setting means. 2. A load prediction type area division method means for setting an allocation method.
The parallel image processing device described. 14. A brightness value calculation formula setting means for setting a calculation formula of a brightness value of a target image, and a brightness value for each pixel of the target image is calculated according to the calculation formula set by the brightness value calculation formula setting means. And a pixel reference load prediction calculation means for calculating a predicted load according to the brightness value calculated by the brightness value calculation execution means and transmitting the predicted load to the load prediction setting means. The parallel image processing device according to claim 13. 15. A pixel reference load prediction condition setting means for setting a calculation method of a prediction load for a luminance value of each pixel, wherein the pixel reference load prediction calculation means is set to the pixel reference load prediction condition setting means. 15. The parallel image processing device according to claim 14, wherein the pixel condition load type pixel reference load prediction calculation means calculates a predicted load for each unit area according to a method. 16. A characteristic region extraction setting means for setting a method of calculating a characteristic region serving as a load prediction reference from the luminance value of a pixel,
And a characteristic region extracting unit that extracts a characteristic region from the luminance value of each pixel of the target image calculated by the luminance value calculation executing unit according to the calculation method set by the characteristic region extraction setting unit. The calculation means is a characteristic area load prediction calculation means for calculating a predicted load for each unit area in the image processing algorithm according to the characteristic area extracted by the characteristic area extraction means and transmitting it to the load prediction setting means. The parallel image processing device according to claim 14, which is characterized in that. 17. A characteristic region load prediction condition setting means for setting a calculation method of a predicted load for a characteristic region, wherein the characteristic region load prediction calculation means is responsive to the calculation method set by the characteristic region load prediction condition setting means. 17. The parallel image processing device according to claim 16, wherein the parallel image processing device is a measurement condition corresponding feature region load prediction calculation unit that calculates a prediction load for each unit area. 18. A reference area setting means for setting an image area in a fixed range as a reference area, wherein the pixel reference load prediction calculation means is a brightness value calculation execution means in the reference area set by the reference area setting means. 15. The parallel image processing device according to claim 14, wherein the reference region load prediction calculation unit calculates a prediction load for each unit region in the image processing algorithm according to the calculated brightness value. 19. A reference area load prediction condition setting means for setting a calculation method of a predicted load according to a brightness value in the reference area, wherein the reference area load prediction calculation means is set by the reference area load prediction condition setting means. 19. The measurement condition-corresponding reference area load prediction calculation means for calculating a predicted load for each unit area according to the calculated calculation method.
The parallel image processing device described. 20. A tilt reference setting means for setting a reference brightness value inclination and a pixel reference load prediction calculation means for predicting each unit area according to the inclination of the brightness value set by the tilt reference setting means. The parallel image processing device according to claim 14, wherein the parallel image processing device is a slope load prediction calculation unit that calculates a load. 21. Slope load prediction condition setting means for setting a calculation method of a prediction load for a slope of a brightness value, wherein the slope load prediction calculation means is set in accordance with the calculation method set by the slope load prediction condition setting means. 21. The parallel image processing apparatus according to claim 20, wherein the parallel load image processing apparatus is a slope load prediction calculation means corresponding to a measurement condition for calculating a prediction load for each unit area. 22. A shooting area load prediction calculation means for calculating a predicted load for each unit area in the image processing algorithm from the area information of the captured image and transmitting the calculated load to the load prediction setting means. Item 14. The parallel image processing device according to item 13. 23. A brightness value calculation formula setting means and a brightness value calculation executing means according to claim 14, a pixel reference load prediction condition setting means and a measurement condition corresponding pixel reference load prediction calculation means according to claim 15, A characteristic region extraction setting unit and a characteristic region extraction unit according to Item 16, a characteristic region load prediction condition setting unit according to Claim 17, and a measurement condition corresponding characteristic region load prediction calculation unit, and a reference region setting unit according to Claim 18. A reference area load prediction condition setting means and a measurement condition corresponding reference area load prediction calculation means according to claim 19, a tilt reference setting means according to claim 20, and a tilt load prediction condition setting means according to claim 21. An inclination load prediction calculation means corresponding to measurement conditions, a shooting area load prediction calculation means according to claim 22, and a prediction calculation weight setting means capable of weighting each prediction condition. 14. The method according to claim 13, further comprising:
The parallel image processing device described. 24. A predictive value recording means for recording the calculated predicted value and the input condition, a processing result recording means for recording a processing load calculated in actual execution for the predicted value, and the predicted value recording means. 24. Any one of claims 14 to 23, further comprising: a prediction parameter correction means for correcting the calculation method of the prediction load based on the recorded prediction value and the processing load recorded by the processing result recording means. Or the parallel image processing device according to item 1. 25. A shooting condition load prediction that records a condition of capturing an image at the time of executing a past prediction calculation and selects an optimum correction of the prediction calculation from a shooting condition for an image for which a new prediction calculation is performed. 25. The parallel image processing device according to claim 24, further comprising a correction unit. 26. When instructed to calculate a new predicted value,
While searching for and selecting a similar condition from among the input conditions for calculating the predicted value recorded in the predicted value recording means, referring to the calculation result recorded in the processing result recording means, correction for the selected predicted calculation is made. 25. The parallel image processing apparatus according to claim 24, further comprising an optimum prediction condition calculation means for performing the calculation. 27. Pre-calculation selection setting means for setting a condition of whether or not to perform the pre-calculation processing, and pre-calculation determination for judging the presence or absence of the pre-calculation processing based on the conditions set by the pre-calculation selection setting means. Means, and a pre-processing load amount calculation means for executing in advance a calculation for obtaining a processing load amount for each unit area targeting the entire image area when instructed by the pre-calculation determination means.
Based on the processing load amount for each unit area pre-calculated by the pre-processing load amount calculating means, a pre-processing compatible image area allocating means for allocating areas so that the processing loads of the respective processors become equal to each other. The parallel image processing apparatus according to claim 1, further comprising: 28. The area division setting means is an image information reference type area division setting means for setting a method of allocating a processing area in consideration of image information, and the image area allocating means is used when re-allocation is requested by the processor. , Image information reference type image area allocating means for reallocating processing areas by referring to image information, and processing load calculating means for calculating image information for each processing unit with respect to the allocated processing areas 2. The parallel image processing device according to claim 1, wherein the parallel image processing device is a type processing load calculating means. 29. The image information reference type image area allocating means,
29. The parallel image processing apparatus according to claim 28, wherein when the processing areas are reallocated by referring to the image information, the number of unit areas is prioritized to determine the processing areas to be reallocated. 30. The image information reference type image area allocating means,
29. The parallel image processing apparatus according to claim 28, wherein when the processing area is reallocated with reference to the image information, the size of the image information is prioritized to determine the processing area to be reallocated. 31. The image information reference type image area allocating means,
When reallocating the processing area with reference to the image information, the processing load of the image processing to be executed is executed based on the obtained image information instead of simply adopting the processing area to be reallocated from the image information. 31. The parallel image processing device according to claim 28, wherein the processing area is calculated, and the processing area to be reallocated is determined based on the processing load. 32. The image information reference type processing load calculation means,
31. The parallel image processing according to claim 28, wherein at the time of initial allocation, the image information calculation and the processing instruction to the image processing execution means are continuously executed. apparatus. 33. When executing an image processing algorithm in which a processing load changes depending on an input image on a parallel computer,
The allocation processing area is dynamically determined, and the processing restriction amount for the allocation processing area is calculated from the maximum processing load amount in the unit area, and is assigned to the algorithm execution means of each processor as a set of the allocation processing area and the processing restriction amount. Along with assigning
An area exceeding the processing limit amount is received from the algorithm executing means of each processor, the area is incorporated into the entire unprocessed area, and again assigned to the algorithm executing means of each processor as a processing area.
For the allocated image area, calculate the processing load in the target image area, instruct execution of processing in the range that does not exceed the specified processing limit amount, and if the processing limit amount is exceeded, limit A parallel image processing method, characterized by requesting reallocation of an over region. 34. The parallel image processing method according to claim 33, wherein the allocation method of the allocation processing area is determined from the maximum processing load amount in the unit area in the image processing algorithm. 35. Measuring the execution time for the maximum load and the execution time for the minimum load in the unit area of the image processing algorithm, and determining the allocation method of the allocation processing area and the calculation method of the processing limit amount from the measured time. 34. The parallel image processing method according to claim 33. 36. The parallel image processing method according to claim 33, wherein an image area of an arbitrary size is designated as a block, and the designated processing is prioritized within the block. 37. When the processing load on a certain processor increases remarkably and the reallocation based on the unit area in the image processing algorithm cannot sufficiently cope with it, the unit area is divided and reallocated, and at the same time in each processor. 34. The parallel image processing method according to claim 33, further comprising the step of determining a designated processing restriction amount and executing the processing in the divided unit areas in consideration of division of the unit areas. 38. The parallel image processing method according to claim 37, wherein whether or not the unit area can be divided is determined based on the number of waiting processors and the amount of the remaining processing area. 39. When a unit area is divided and a reallocation is requested, the processor which processes the unit area first records the calculation result of the first half performed for obtaining the processing load, and the divided area is divided. 38. The parallel image according to claim 37, wherein a processor instructed later to perform the remaining processing in the unit area uses the recorded calculation result of the first half to perform the second half of the processing. Processing method. 40. The parallel image processing method according to claim 37, wherein when the unit area is divided, a division position at which cache access is advantageous is obtained from the sizes of the cache and the image area. 41. The parallel image processing method according to claim 33, wherein a prediction load is set for each unit area in the image processing algorithm, and the image area allocation method is determined based on the prediction load. 42. The parallel image processing method according to claim 41, wherein the prediction load is calculated based on the luminance value of each pixel of the processing target image. 43. The parallel image processing method according to claim 41, wherein a characteristic region is determined from a luminance value of each pixel of the image to be processed, and the prediction load is calculated based on the characteristic region. 44. The parallel image processing method according to claim 41, wherein a predetermined range of image areas is set as a reference area, and the prediction load is calculated using the luminance value of each pixel in the reference area. . 45. The calculated predicted value and the input condition for the predicted value are recorded, the processing load calculated in actual execution is recorded for the predicted value, and the predicted load calculation method is based on these recorded values. The correction is performed.
1. The parallel image processing method described in 1. 46. A method for newly calculating a predicted value is selected from the conditions for capturing an image at the time of executing the past prediction calculation and the input condition used as a calculation standard for the predicted value, and an appropriate correction is made. 46. The parallel image processing method according to claim 45, which is performed. 47. It is determined whether or not a processing load for each unit area is obtained in advance, and when the processing load for each unit area is obtained in advance, an image area is allocated based on the obtained load. 34. The parallel image processing method according to claim 33, wherein a method is determined.