JP5708303B2 - Image processing apparatus, image processing method, and image processing program - Google Patents

Description

  The present invention relates to an image processing apparatus and the like.

  A vehicle-mounted device having a monitor, such as a car navigation system, includes a drawing processor for drawing an image on the monitor. The drawing processor includes a plurality of drawing sub-processors. The drawing processor sequentially assigns processing to the drawing sub-processor and processes all pixels of the image data.

  The drawing sub-processor performs an operation for determining one assigned output pixel value, and is released after the operation is completed. The drawing subprocessor performs an operation using each element constituting the pixel value as one vector. For example, each element includes R (Red), G (Green), B (Blue), and A (Alpha) elements.

  Some in-vehicle devices provide driver assistance by performing image analysis in addition to the above processing. This image analysis includes filter processing such as smoothing and edge detection for the image data.

  The filter processing is desirably performed at high speed in order to realize real-time processing. For this reason, there is a technique for performing filter processing using a drawing processor capable of parallel processing. In this technique, camera video is converted into one-element image data and analyzed. For example, one element of image data corresponds to a grayscale image composed of luminance values.

  Here, if a drawing sub-processor that performs a vector operation on a plurality of elements directly performs processing on image data of one element, the calculation cannot be performed efficiently. For this reason, in the prior art, one element of image data is divided by the number of n elements that can be processed in parallel within the drawing sub-processor, and each drawing sub-processor performs parallel filtering on the n-divided image data. Run. For example, n is a natural number of 2 or more.

JP 2007-274478 A JP 2010-98633 A

  However, the above-described conventional technique has a problem that the filter processing for image data cannot be accelerated.

  When the filter processing is performed by dividing one element of image data as in the prior art, the processing for dividing the image data, the filtering processing for each image data, and the processing for combining the divided image data are performed. . For this reason, it is advantageous in terms of speed to perform the filtering process on the image data without dividing the image data because the dividing process and the combining process do not have to be performed.

  Note that the above problem is not limited to the in-vehicle device, and similarly occurs in an image processing device that performs filter processing on image data.

  The disclosed technique has been made in view of the above, and an object thereof is to provide an image processing apparatus, an image processing method, and an image processing program capable of speeding up a filtering process on image data.

  The disclosed image processing apparatus includes an evaluation unit and a filter processing unit. The evaluation unit evaluates the calculation amount of various division methods based on the size of the image data including fewer elements than the plurality of elements and the filter size. The filter processing unit executes filter processing based on the evaluation result of the evaluation unit.

  According to the disclosed image processing apparatus, it is possible to increase the speed of the filtering process on the image data.

FIG. 1A is a diagram for explaining image processing by a drawing processor. FIG. 1B is a diagram for explaining image processing by the drawing processor. FIG. 2 is a diagram illustrating an example of filter processing. FIG. 3 is a diagram (1) for supplementarily explaining the size of the filter. FIG. 4 is a diagram (2) for supplementarily explaining the size of the filter. FIG. 5 is a functional block diagram of the configuration of the image processing apparatus according to the present embodiment. FIG. 6 is a diagram (1) illustrating an example of the division method. FIG. 7 is a diagram (2) illustrating an example of the division method. FIG. 8 is a diagram (3) illustrating an example of the division method. FIG. 9 is a diagram for explaining the application range of the filter. FIG. 10 is a diagram illustrating an example of a data structure of parameters. FIG. 11A is a diagram (1) for supplementarily explaining the evaluation function. FIG. 11B is a diagram (2) for supplementarily explaining the evaluation function. FIG. 11C is a diagram (3) for supplementarily explaining the evaluation function. FIG. 11D is a diagram (4) for supplementarily explaining the evaluation function. FIG. 12 is a diagram illustrating a calculation amount for each division method. FIG. 13 is a functional block diagram illustrating the configuration of the filter processing unit. FIG. 14 is a diagram for explaining the processing of the dividing unit. FIG. 15 is a diagram illustrating the relationship between the start coordinates of the divided image data and the offset. FIG. 16 is a flowchart of the process procedure of the image processing apparatus according to the present embodiment. FIG. 17 is a diagram illustrating an example of a computer that executes an image processing program.

  Embodiments of an image processing apparatus, an image processing method, and an image processing program disclosed in the present application will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments.

  Before describing the image processing apparatus according to the present embodiment, image processing by a drawing processor used by the image processing apparatus will be described. 1A and 1B are diagrams for explaining image processing of a drawing processor. As illustrated in FIG. 1A, the drawing processor 10a includes drawing sub-processors 1b to 9b that perform vector operations.

  The drawing processor 10a sequentially assigns processes to the drawing sub-processors 1b to 9b, and determines all pixel values of the output image 10c. The drawing sub-processors 1b to 9b perform an operation for determining a pixel value for one allocated coordinate, and are released after completion.

  For example, the drawing sub-processor 1b is assigned to the coordinate 1d = (x1, y1) in FIG. 1B. In addition, the drawing sub-processor 2b is assigned to the coordinates 2d = (x2, y2). A drawing sub-processor 3b is assigned to the coordinates 3d = (x3, y3). Each drawing sub-processor can perform similar vector operations on a plurality of elements. The vector operation of each drawing sub-processor is executed independently.

  In FIG. 1B, the output image 10c is composed of four elements 1c to 4c, and each pixel element is processed by one vector operation instruction. For example, when the drawing sub-processor 1b generates output image data in which the pixel value of the coordinate 1d of the input image data is doubled, the pixel value is doubled in the common coordinate 1d of 1c to 4c.

  Next, the number of elements will be described. For example, image data composed of one element is an image expressed by a luminance value. Also, RGB image data having a color has three elements of R, G, and B. In addition, image data having a color may have four elements of R, G, B, and A. This “A” is an alpha element in consideration of superposition with other image data.

  Next, an example of filter processing performed by the image processing apparatus will be described. In the filter processing, the pixel value of the target pixel after the filter processing is calculated by inputting the pixel value around the target pixel and the target pixel, accumulating correspondingly called arrays called filter coefficients, and calculating an accumulated value.

  FIG. 2 is a diagram illustrating an example of filter processing. Reference numeral 20a in FIG. 2 denotes image data to be filtered. The image data 20a has nine pixels corresponding to the coordinates (x1, y1) to (x3, y3). The numbers included in the pixels at each coordinate correspond to the pixel values. For example, the pixel value of the image at the coordinates (x1, y1) is “10”. 2b in FIG. 2 is an example of the data structure of the filter.

  Here, as an example, a case where the pixel value of the coordinates (x2, y2) of the image data 20a is obtained by applying the filter 20b to the image data 20a will be described. The pixel value of the coordinates (x2, y2) is obtained by the following equation (1), and the pixel value after the filter processing is “50”. Similarly, the filter 20b is applied to pixels having other coordinates. Thus, the filter processing of the image data 20a is performed by applying the filter 20b to the coordinates (x1, y1) to (x3, y3).

Pixel value of coordinates (x2, y2) = 10 × 1/9 + 20 × 1/9 + 30 × 1/9 + 40 × 1/9 + 90 × 1/9 + 60 × 1/9 + 70 × 1/9 + 80 × 1/9 + 50 × 1/9. (1)

  A filter that calculates an average value including pixel values of surrounding pixels, such as the filter shown in FIG. 2, may be called a smoothing filter. This smoothing filter is used when noise included in an image is removed.

There are filters of various sizes depending on the purpose. In this embodiment, the size of a filter composed of vertical and horizontal rectangles is represented by (f _w , f _h ). Note that the values of f _w and f _h are expressed by a value obtained by subtracting one pixel from the number of pixels from the central pixel to one end. 3 and 4 are diagrams for supplementarily explaining the size of the filter. For example, the size (f _w1 , f _h1 ) of the filter 21a shown in FIG. 3 is (2, 2). The size (f _w2 , f _h2 ) of the filter 21b shown in FIG. 4 is (3, 1).

  Next, the configuration of the image processing apparatus according to the present embodiment will be described. FIG. 5 is a functional block diagram of the configuration of the image processing apparatus according to the present embodiment. As illustrated in FIG. 5, the image processing apparatus 100 includes an input unit 110, an output unit 120, a storage unit 130, and a control unit 140.

  The input unit 110 is an input device that inputs image data or parameters. Here, the image data is image data to which the filtering process is applied. The image data is one-element image data. For example, the image data corresponds to a gray scale image represented by a luminance value. In the present embodiment, as an example, image data of one element is used. However, the present invention is not limited to this, and image data having a smaller number of elements than the number of vector operations may be used.

  The parameter is data used when calculating the amount of calculation when the filtering process is executed by various methods. For example, this parameter includes the size of image data to be output as a filter processing result, the size of the filter, the time for inputting and outputting pixel values to the register of the rendering processor, and the time required for the filter processing. Details of the parameters will be described later.

  The output unit 120 is an output device that outputs image data that has been filtered by the control unit 140.

  The storage unit 130 stores image data 131, parameters 132, divided image data 133, filter data 134, and composite image data 135. The storage unit 130 corresponds to, for example, a semiconductor memory device such as a random access memory (RAM), a read only memory (ROM), and a flash memory, or a storage device such as a hard disk or an optical disk.

  The image data 131 corresponds to the image data input from the input unit 110. The parameter 132 corresponds to a parameter input from the input unit 110.

  The divided image data 133 corresponds to the image data divided by the control unit 140. The filter data 134 is image data that is the result of performing a filtering process on the image data. The composite image data 135 is image data obtained by combining the divided image data 133.

  The control unit 140 includes an image data reception unit 141, a parameter reception unit 142, an evaluation unit 143, and a filter processing unit 144. The control unit 140 corresponds to an integrated device such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array). The control unit 140 corresponds to an electronic circuit such as a CPU or MPU (Micro Processing Unit). The control unit 140 includes a processor that executes vector operations.

  The image data receiving unit 141 acquires image data from the input unit 110 and stores the acquired image data in the storage unit 130. The parameter receiving unit 142 acquires parameters from the input unit 110 and causes the storage unit 130 to store the acquired parameters.

  The evaluation unit 143 is a processing unit that evaluates the calculation amount when the filter processing is executed based on the parameter 132 for each of various division methods. The evaluation unit 143 outputs the evaluation result to the filter processing unit 144. Details of the processing of the evaluation unit 143 will be described later.

  The filter processing unit 144 is a processing unit that divides the image data 131 based on the evaluation result of the evaluation unit 143 by a division method that requires the least amount of calculation for the filter process and executes the filter process. The filter processing unit 144 outputs the processing result to the output unit 120. Details of the processing of the filter processing unit 144 will be described later.

  Next, a division method when the filter processing unit 144 divides the image data 131 will be described. For example, when the drawing processor used by the image processing apparatus 100 can execute a vector calculation of four elements, the vector calculation can be efficiently executed by dividing the image data 131 into four.

  The method of dividing the image data into four is not one, but there are three methods as shown in FIGS. 6, 7 and 8 are diagrams illustrating an example of the division method. Here, in order to distinguish each division method, the division method of 22a shown in FIG. 6 is 4 × 1 division, the division method of 22b shown in FIG. 7 is 1 × 4 division, and the division method of 22c shown in FIG. Is divided into 2 × 2.

  In this embodiment, a method of not dividing image data is one of the division methods. In this way, the method of not dividing the image data is set as non-division. For this reason, there are four types of division methods of the present embodiment: 4 × 1 division, 1 × 4 division, 2 × 2 division, and non-division.

  Depending on each division method, the amount of calculation when performing the filter processing is different. The reason why the calculation amount is different will be described below. In general-purpose two-dimensional filter processing, a filter is applied to pixels other than the end pixel of the image data. This is because if a filter is applied to the pixels at the edge of the image data, part of the filter protrudes from the image data.

  FIG. 9 is a diagram for explaining the application range of the filter. The image data 30a in FIG. 9 is image data before division. For example, when the image data 30a is divided by 2 × 2, the divided image data 31a to 34a are obtained.

  As described above, the applicable range of the filter processing is other than the pixels at the end portion of the image data. For example, if the filter 30c is as shown in FIG. 6, the filter cannot be applied to pixels above the coordinates 30b. If a filter is applied to a pixel above the coordinate 30b, the filter 30c protrudes from the image data 30a. For this reason, in the image data 30a, the range that can be filtered is the inside surrounded by the shaded portion.

  Then, if the image data 30a is divided as it is, the range in which the filter processing can be performed is further narrowed. For this reason, it is preferable to provide an overlapping region so that a filter can be applied to pixel values near the division boundary. This overlapping region corresponds to an image in a region that protrudes by half the filter size from the pixel at the boundary portion of the divided image data.

  In FIG. 9, the divided image data 31a has an overlapping area 31d. This overlapping area 31d corresponds to an area in which the size of the filter is included in the boundary portion of the divided image data 31a. For example, the overlapping area 31d is data near the boundary between the divided image data 32a and 33a adjacent to the divided image data 31a. In this way, by providing the divided image data 31a with the overlapping region 31d, the filter 30f can be applied to the pixel value of the coordinate 30e of the boundary portion.

Similarly, the divided image data 32a has an overlapping region 32d. The divided image data 33a has an overlapping area 33d. The divided image data 34a has an overlapping area 34d. Thus, even if the image data 30a is divided by providing the overlapping regions 32d to 34d,
The scope of filter processing is not narrowed.

  For this reason, when dividing one element of image data and performing a vector operation, an image dividing process, a filter process, and an image synthesizing process are sequentially executed. In the process of dividing the image, the image data is divided in consideration of the overlapping area. On the other hand, if the filter processing is performed as it is without dividing the image data, the processing for dividing the image and the processing for synthesizing the image need not be executed.

  Next, the process of the evaluation unit 143 will be described in detail. The evaluation unit 143 calculates the calculation amount T of the filter processing by each division method using the evaluation function of Expression (2). In Expression (2), “D” is a calculation amount for creating divided image data from image data. “F” is a calculation amount of the filter processing. “C” is a calculation amount of the process of combining the divided image data after the filter process. Each calculation amount corresponds to a processing time.

  T = D + F + C (2)

Before describing the details of the evaluation function, the parameter 132 for calculating the calculation amount T will be described. FIG. 10 is a diagram illustrating an example of a data structure of parameters. As shown in FIG. 10, the parameter 132 includes s _w , s _h , f _w , f _h , p _i , p _o , f ₁ , and f ₂ . Among these, “s _w ” corresponds to the horizontal size of the image output as the filter process. “S _h ” corresponds to the vertical size of the image output as the filter process.

“F _w ” is the horizontal size of the filter. “F _h ” is the vertical size of the filter. However, as already described, the values of f _w and f _h are expressed by a value obtained by subtracting one pixel from the number of pixels from the central pixel to one end.

“P _i ” is the time for transferring the pixel value to the register of the drawing processor. “P _o ” is the time for transferring the pixel value from the register of the drawing processor for output. In the case of a drawing processor capable of four-element vector calculation, in the input / output transfer with respect to the same coordinates, the transfer of only one element and the transfer of four elements can be processed with p _i and p _o at the same time. is there.

“F ₁ ” is a filter processing time depending on the filter size. “F ₂ ” is a filter processing time independent of the filter size.

The parameter receiving unit 142, the parameter value of _{"s w, s h, f w} , f h, p i, p o, f 1, f 2 ", may be obtained by any method. For example, the parameter receiving unit 142 the value of "s _w, s _h", is acquired from the image size of the applications that require filtering. The parameter receiving unit 142 acquires the values of “f _w , f _h ” from the filter size of the target filter process.

The parameter receiving unit 142 acquires the values of “p _i , p _o ” by executing the drawing processor specifications and a simple benchmark. For example, the parameter receiving unit 142 sets the value of “p _i , p _o ” as the number of cycles for acquiring the unit pixel of the drawing processor.

The parameter receiving unit 142 acquires the values of “f ₁ , f ₂ ” by performing a benchmark for each element for a part of the target filter processing and executing a static analysis of the code. Note that the parameters “p _i , p _o , f ₁ , f ₂ ” need only have the same time unit, and a scale other than the number of cycles may be used. In addition, as the values of the parameters, those investigated by the administrator in advance may be used.

  Subsequently, the expression (2) will be described separately for “in the case of non-division” and “in the case of 4 × 1 division, 1 × 4 division, and 2 × 2 division”.

  First, in the case of non-division, the divided image data is not generated, and the process of combining the divided images is not executed. Note that “F” in Expression (2) is expressed by Expression (3).

On the other hand, in the case of “4 × 1 division, 1 × 4 division, 2 × 2 division”, “D”, “F”, and “C” in Expression (2) are respectively , (4), (5), and (6). In Expressions (4), (5), and (6), d _w and d _h are the number of horizontal divisions and the number of vertical divisions at the time of image division. For example, in the case of 4 × 1 division, the value of d _w is 4 and the value of d _h is 1. As the values of d _w and d _h , a combination of values set in advance from the number of elements of vector operations of the drawing processor may be used.

11A to 11D are diagrams for supplementarily explaining the evaluation function. The description regarding s _w , s _h , f _w , and f _h included in FIGS. 11A to 11D is the same as that described with reference to FIG. As shown in FIG. 11A, when the size of the filter 41a is expressed using f _w and f _h , the horizontal width of the filter is “2f _w +1” and the vertical width is “2f _h +1”.

  When the image data 40a shown in FIG. 11A is divided 4 × 1, four pieces of divided image data 40b shown in FIG. 11B are generated. Further, when the image data 40a shown in FIG. 11A is divided into 1 × 4, four pieces of divided image data 40c shown in FIG. 11C are generated. Further, when the image data 40a shown in FIG. 11A is divided by 2 × 2, four pieces of divided image data 40d shown in FIG. 11D are generated.

A supplementary explanation will be given for equation (4). When the image data 40a is divided, one of the divided image data 40b to 40d is generated. Here, the time for obtaining pixel values of d _w × d _h different coordinates from the image data of one element and outputting d _w × d _h image values is “p _i × d _w × d _h + p _o ”. For this reason, at the time of division, the amount of calculation for creating divided image data from the image data can be expressed by the above equation (4).

  A supplementary explanation will be given of equations (3) and (5). The product of the time required to output one coordinate and the size of the image data to be processed corresponds to the time for the filter processing. The time required to output one coordinate is the sum of the time dependent on the filter size including the input, the time independent of the filter size, and the output time of the processing result. In the time required for outputting one coordinate, there is no difference between division and non-division.

Multiplying the time required to output one coordinate by the screen size corresponds to the filter processing time. The processing screen size at the time of non-division is “s _w × s _h ”. The processing screen size at the time of division is “s _w × s _h / d _w / d _h ”.

A supplementary explanation will be given for equation (6). The process of combining the divided image data requires a processing time for performing image conversion of one element from an image of d _w × d _h elements. In order to determine the output values for one coordinate with the image data of one element, a coordinate of the image data d _w × d _h elements read by p _i, outputs at time p _o. This process will be carried out for the size s _w × s _h of the image data after output. For this reason, the calculation amount of the process of combining the divided image data is expressed by Expression (6).

  The evaluation unit 143 calculates the amount of calculation for each division method based on the above equations (2) to (6) and the parameter 132. FIG. 12 is a diagram illustrating a calculation amount for each division method. In the example shown in FIG. 12, the calculation amount of each division method for the parameters 50a to 50f is shown.

For example, the value of the parameter 50a is “s _w = 636”, “s _h = 476”, “f _w = 2”, “f _h = 2”, “p _i = 200”, “p _o = 200”, “ f ₁ = 20 ”and“ f ₂ = 20 ”. The value of the parameter 50b is “s _w = 636”, “s _h = 480”, “f _w = 2”, “f _h = 0”, “p _i = 200”, “p _o = 200”, “f ₁ = 20 "and" f ₂ = 20 ". The value of the parameter 50c is “s _w = 640”, “s _h = 476”, “f _w = 0”, “f _h = 2”, “p _i = 200”, “p _o = 200”, “f ₁ = 20 "and" f ₂ = 20 ".

The value of the parameter 50d is “s _w = 638”, “s _h = 480”, “f _w = 1”, “f _h = 0”, “p _i = 50”, “p _o = 400”, “f ₁ = 20 "and" f ₂ = 20 ". The value of the parameter 50e is “s _w = 638”, “s _h = 480”, “f _w = 1”, “f _h = 0”, “p _i = 50”, “p _o = 400”, “f ₁ = 100 "and" f ₂ = 50 ". The value of the parameter 50f is “s _w = 638”, “s _h = 480”, “f _w = 1”, “f _h = 0”, “p _i = 60”, “p _o = 60”, “f ₁ = 20 "and" f ₂ = 20 ".

  The calculation amount of each division method using the parameter 50a is “632.3” for 4 × 1 division, “632.7” for 1 × 4 division, “631.9” for 2 × 2 division, and “631.9” for non-division. 1731.7 ". For this reason, in the case of the parameter 50a, the calculation amount of 2 × 2 division is minimized.

  The calculation amount of each division method based on the parameter 50b is “301.1” for 4 × 1 division, “299.7” for 1 × 4 division, “300.1” for 2 × 2 division, and “403. 0 ". For this reason, in the case of the parameter 50b, the calculation amount of 1 × 4 division is minimized.

  The calculation amount of each division method based on the parameter 50c is “299.2” for 4 × 1 division, “301.1” for 1 × 4 division, “299.8” for 2 × 2 division, and “402. 1 ". For this reason, in the case of the parameter 50c, the calculation amount of 4 × 1 division is minimized.

  The calculation amount of each division method based on the parameter 50d is “232.6” for 4 × 1 division, “232.1” for 1 × 4 division, “232.2” for 2 × 2 division, and “193. 9 ". For this reason, in the case of the parameter 50d, the undivided calculation amount is minimized.

  The calculation amount of each division method using the parameter 50e is “253.2” for 4 × 1 division, “252.8” for 1 × 4 division, “252.9” for 2 × 2 division, and “275. 6 ”. For this reason, in the case of the parameter 50e, the calculation amount of 1 × 4 division is minimized.

  The calculation amount of each division method by the parameter 50f is “84.5” for 4 × 1 division, “84.3” for 1 × 4 division, “84.4” for 2 × 2 division, and “98. 0 ". For this reason, in the case of the parameter 50f, the calculation amount of 1 × 4 division is minimized.

  The evaluation unit 143 outputs the calculation amount information of each division method calculated based on the parameters to the filter processing unit 144.

  Next, the filter processing unit 144 will be described in detail. FIG. 13 is a functional block diagram illustrating the configuration of the filter processing unit. As shown in FIG. 13, the filter processing unit 144 includes a dividing unit 144a, a filter unit 144b, a combining unit 144c, and an output unit 144d.

  The dividing unit 144a is a processing unit that divides the image data 131 based on the evaluation result from the evaluation unit 143. The dividing unit 144a notifies the type of the executed division method to the filter unit 144b, the combining unit 144c, and the output unit 144d.

  For example, the dividing unit 144a notifies the filter unit 144b that the division is not performed when the calculation amount of “non-dividing” is the smallest. When the calculation amount of “4 × 1 division” is minimum, the dividing unit 144a notifies the filter unit 144b that the 4 × 1 division has been performed. When the calculation amount of “1 × 4 division” is minimum, the dividing unit 144a notifies the filter unit 144b that 1 × 4 division has been performed. When the calculation amount of “2 × 2 division” is minimum, the dividing unit 144a notifies the filter unit 144b that 2 × 2 division has been performed.

  Processing when the dividing unit 144a divides the image data 131 will be described. Here, as an example, a case where the dividing unit 144a acquires the evaluation result of the parameter 50a in FIG. In the evaluation result of the parameter 50a, the one with the smallest calculation amount is “2 × 2 division”. Therefore, the dividing unit 144a selects “2 × 2 division” and divides the image data 131.

FIG. 14 is a diagram for explaining the processing of the dividing unit. As shown in FIG. 14, the dividing unit 144a divides the image data 131 into divided image data 131a, 131b, 131c, and 131d. As shown in FIG. 14, the divided image data 131a to 131d include overlapping areas 60a to 60d, respectively. The _{s w} of the image data 131 "636", when the _{s h} and "476", the horizontal size w of the divided image data 131a~131d including overlap region becomes "322", the vertical size h is "242". This, w is a s _w to the value in half "636" is equal to the sum of the values of the filter size f _w × 2. h is equal to a value obtained by halving “476” and adding a value of the filter size f _w × 2.

FIG. 15 is a diagram illustrating the relationship between the start coordinates of the divided image data and the offset. The coordinates obtained by adding the start coordinates and the offset are the actual start coordinates of the divided image data 131a to 131d. The offset is set according to the filter sizes f _w and f _h . For example, when the image data 131 is divided as it is, each divided image data does not include an overlapping image, but the start coordinates of the divided image data including the overlapping image can be designated by considering an offset.

  The actual start coordinate of the divided image data 131b is (318, 0). This coordinate (318, 0) is the upper left coordinate of the divided image data 131b in FIG. The actual start coordinates of the divided image data 131c are (0, 238). The coordinates (0, 238) are the upper left coordinates of the divided image data 131c in FIG. The actual start coordinates of the divided image data 131d are (318, 238). These coordinates (318, 238) are the upper left coordinates of the divided image data 131d in FIG.

  The dividing unit 144a causes the storage unit 130 to store the divided image data 131 as the divided image data 133. The divided image data 133 corresponds to the divided image data 131a to 131d in FIG.

  The filter unit 144b is a processing unit that performs the filtering process illustrated in FIG. 2 on the image data.

  A process of the filter unit 144b when the image data 131 is divided into the division unit 144a will be described. The filter unit 144b performs a filtering process on the divided image data 133, and stores the filtered divided image data in the storage unit 130 as the filter data 134. When the filter processing is executed on the divided image data 131a to 131d shown in FIG. 14, the internal size 318 × 238 of 322 × 242 is the output size of the filter processing.

  A process of the filter unit 144b when the image data 131 is not divided into the division unit 144a will be described. The filter unit 144 b performs a filtering process on the image data 131 and causes the storage unit 130 to store the filtered image data as filter data 134. When the filter processing is executed on the image data 131 shown in FIG. 14, the internal size of 640 × 480 636 × 476 is the output size of the filter processing.

  The combining unit 144c is a processing unit that combines the divided image data. Note that the combining unit 144c executes processing when the dividing unit 144a divides the image data 131. The synthesizing unit 144c acquires each divided image data subjected to the filter processing from the filter data 134, and causes the storage unit 130 to store the synthesized image of the divided image data as the synthesized image data 135.

  For example, for the divided image data 131a to 131d after the filter processing, the synthesis unit 144c designates the start coordinate + offset shown in FIG. 15 as the start coordinate of the synthesis destination, and sets the effective area of each element to 318 × 238. By executing such processing by the synthesizing unit 144c, each divided image data becomes original image data of size 636 × 476.

  The output unit 144d is a processing unit that outputs the image data after the filter processing. When the image data 131 is not divided into the dividing unit 144a, the output unit 144d outputs the filter data 134. On the other hand, when the image data 131 is divided into the dividing unit 144a, the composite image data 135 is output.

  Next, a processing procedure of the image processing apparatus 100 according to the present embodiment will be described. FIG. 16 is a flowchart of the process procedure of the image processing apparatus according to the present embodiment. For example, the process shown in FIG. 16 is executed in response to the acquisition of image data to be filtered.

  As shown in FIG. 16, the image processing apparatus 100 acquires image data (step S101) and evaluates the calculation amount of each division method (step S102). The image processing apparatus 100 determines whether to divide the image data (step S103).

  When the image processing apparatus 100 does not divide the image data (No at Step S103), the image processing apparatus 100 proceeds to Step S105. On the other hand, when the image processing apparatus 100 divides the image data (step S103, Yes), the image processing apparatus 100 divides the image data (step S104).

  The image processing apparatus 100 executes filter processing (step S105), and determines whether or not the image data has been divided (step S106). When the image processing apparatus 100 has not divided the image data (No at Step S106), the image processing apparatus 100 proceeds to Step S108. When the image processing apparatus 100 divides the image data (Yes at Step S106), the image processing apparatus 100 combines the image data (Step S107) and outputs the image data subjected to the filter processing (Step S108).

  Next, effects of the image processing apparatus 100 according to the present embodiment will be described. The image processing apparatus 100 evaluates the calculation amount of the process in which the evaluation unit 143 divides and filters the image and the calculation amount in the case of performing the filter process without dividing the image, and the filter processing unit 144 performs the division with a small calculation amount. Perform filtering by the method. For this reason, according to the image processing apparatus 100, it is possible to speed up the filter processing executed by the processor capable of vector calculation.

  In addition, the image processing apparatus 100 evaluates the amount of calculation when the divided image data is subjected to the filter processing in parallel, including the process of adding the overlapping area corresponding to the filter size to the divided image data. . For this reason, according to the image processing apparatus 100, the calculation amount of each division method can be more accurately evaluated.

  Further, when dividing the image data, the image processing apparatus 100 specifies the number of image data to be divided according to the number of vector operations. For this reason, according to the image processing apparatus 100, the calculation resource by vector calculation can be utilized without waste.

  Further, the image processing apparatus 100 calculates the amount of calculation for each division method based on the size of the image data, the input / output time of the pixel value to the processor, the time required for the filtering process, and the number of divisions of the image data. Therefore, according to the image processing apparatus 100, it is possible to accurately evaluate the calculation amount of each division method.

  Next, an example of a computer that executes an image processing program that realizes the same function as that of the image processing apparatus 100 illustrated in FIG. 5 will be described. FIG. 17 is a diagram illustrating an example of a computer that executes an image processing program.

  As illustrated in FIG. 17, the computer 200 includes a CPU 201 that executes various arithmetic processes, an input device 202 that receives data input from a user, and a display 203. The computer 200 includes a reading device 204 that reads a program and the like from a storage medium, and an interface device 205 that exchanges data with other computers via a network. The computer 200 also includes a RAM 206 that temporarily stores various information and a hard disk device 207. The devices 201 to 207 are connected to the bus 208.

  The hard disk device 207 stores an evaluation program 207a and a filter processing program 207b. The CPU 201 reads out the evaluation program 207 a and the filter processing program 207 b and develops them in the RAM 206. The evaluation program 207a functions as an evaluation process 206a. The filter processing program 207b functions as a filter processing process 206b.

  For example, the evaluation process 206a corresponds to the evaluation unit 143 in FIG. The filter process 206b corresponds to the filter processing unit 144 in FIG.

  Note that the programs 207a and 207b are not necessarily stored in the hard disk device 207 from the beginning. For example, each program is stored in a “portable physical medium” such as a flexible disk (FD), a CD-ROM, a DVD disk, a magneto-optical disk, and an IC card inserted into the computer 200. Then, the computer 200 may read and execute the programs 207a and 207b from these.

  The following supplementary notes are further disclosed with respect to the embodiments including the above examples.

(Supplementary Note 1) When a processor that performs vector operation on image data including a plurality of elements executes filter processing on image data including fewer elements than the plurality of elements,
Based on the size of the image data including the elements smaller than the plurality of elements and the filter size applied in the filtering process, the filtering process is executed in parallel on each divided image data obtained by dividing the image data. An amount of calculation for each of the plurality of division methods, and an evaluation unit for evaluating the calculated amount of calculation and the amount of calculation in the case of executing the filter processing without dividing the image data,
An image processing apparatus comprising: a filter processing unit that executes the filter processing based on an evaluation result of the evaluation unit.

(Additional remark 2) The said evaluation part performs this filter process with respect to each divided | segmented image data including the process which provides the duplication area | region according to the said filter size to the image data containing an element smaller than the said some element. The image processing apparatus according to appendix 1, wherein the amount of calculation when executed in parallel is evaluated.

(Additional remark 3) When the said filter process part divides | segments the image data containing an element smaller than the said several element, it specifies the division | segmentation number of this image data according to the said vector operation number, It is characterized by the above-mentioned. The image processing apparatus according to appendix 1 or 2.

(Additional remark 4) The said evaluation part is based on the size of the image data containing an element smaller than the said several elements, the input / output time of the pixel value with respect to a processor, the time which the said filter process requires, and the division | segmentation number of this image data The image processing apparatus according to appendix 1, 2, or 3, wherein the amount of calculation is evaluated.

(Supplementary Note 5) An image processing method executed by a computer,
When a processor that performs a vector operation on image data including a plurality of elements executes filter processing on image data that includes fewer elements than the plurality of elements,
Based on the size of the image data including the elements smaller than the plurality of elements and the filter size applied in the filtering process, the filtering process is executed in parallel on each divided image data obtained by dividing the image data. Calculating the amount of calculation for each of the plurality of division methods, and evaluating the calculated amount of calculation and the amount of calculation in the case of executing the filter processing without dividing the image data,
An image processing method characterized by causing each process to execute a filter process based on an evaluation result.

(Additional remark 6) The process which evaluates the said calculation amount is included with respect to each divided | segmented image data including the process which provides the duplication area | region according to the said filter size to the image data containing fewer elements than the said some element The image processing method according to appendix 5, wherein the amount of calculation when the filter processing is executed in parallel is evaluated.

(Additional remark 7) The process which performs the said filter process specifies the division | segmentation number of this image data according to the said vector operation number, when dividing | segmenting the image data containing an element smaller than the said several element 7. The image processing method according to appendix 5 or 6, which is a feature.

(Supplementary Note 8) The processing for evaluating the calculation amount includes the size of image data including elements smaller than the plurality of elements, the input / output time of pixel values for the processor, the time required for the filtering process, and the number of divisions of the image data The image processing method according to appendix 5, 6 or 7, wherein the amount of calculation is evaluated based on the above.

(Appendix 9)
When a processor that performs a vector operation on image data including a plurality of elements executes filter processing on image data that includes fewer elements than the plurality of elements,
Based on the size of the image data including the elements smaller than the plurality of elements and the filter size applied in the filtering process, the filtering process is executed in parallel on each divided image data obtained by dividing the image data. Calculating the amount of calculation for each of the plurality of division methods, and evaluating the calculated amount of calculation and the amount of calculation in the case of executing the filter processing without dividing the image data,
An image processing program for executing each process for executing a filter process based on an evaluation result.

(Additional remark 10) The process which evaluates the said calculation amount is included with respect to each divided | segmented image data including the process which provides the duplication area | region according to the said filter size to the image data containing fewer elements than the said some element The image processing program according to appendix 9, wherein the amount of calculation when the filter processing is executed in parallel is evaluated.

(Additional remark 11) The process which performs the said filter process specifies the division | segmentation number of this image data according to the said vector operation number, when dividing | segmenting the image data containing an element smaller than the said several element The image processing program according to appendix 9 or 10, which is a feature.

(Supplementary Note 12) The processing for evaluating the calculation amount includes the size of image data including elements smaller than the plurality of elements, the input / output time of pixel values to the processor, the time required for the filtering process, and the number of divisions of the image data The image processing program according to appendix 9, 10 or 11, wherein the calculation amount is evaluated based on

DESCRIPTION OF SYMBOLS 100 Image processing apparatus 110 Input part 120 Output part 130 Storage part 140 Control part

Claims (6)

When a processor that performs a vector operation on image data including a plurality of elements executes filter processing on image data that includes fewer elements than the plurality of elements,
Based on the size of the image data including the elements smaller than the plurality of elements and the filter size applied in the filtering process, the filtering process is executed in parallel on each divided image data obtained by dividing the image data. An amount of calculation for each of the plurality of division methods, and an evaluation unit for evaluating the calculated amount of calculation and the amount of calculation in the case of executing the filter processing without dividing the image data,
An image processing apparatus comprising: a filter processing unit that executes the filter processing based on an evaluation result of the evaluation unit.   The evaluation unit executes the filter process in parallel on each divided image data including a process of adding an overlapping area according to the filter size to image data including elements smaller than the plurality of elements. The image processing apparatus according to claim 1, wherein a calculation amount in the case is evaluated. The filter processing unit, when dividing image data including fewer elements than the plurality of elements, specifies the number of divisions of the image data in accordance with the number of elements on which the vector operation is performed. The image processing apparatus according to claim 1.   The evaluation unit calculates the amount of calculation based on the size of image data including elements smaller than the plurality of elements, the input / output time of pixel values to the processor, the time required for the filtering process, and the number of divisions of the image data. The image processing apparatus according to claim 1, wherein evaluation is performed. An image processing method executed by a computer,
When a processor that performs a vector operation on image data including a plurality of elements executes filter processing on image data that includes fewer elements than the plurality of elements,
Based on the size of the image data including the elements smaller than the plurality of elements and the filter size applied in the filtering process, the filtering process is executed in parallel on each divided image data obtained by dividing the image data. Calculating the amount of calculation for each of the plurality of division methods, and evaluating the calculated amount of calculation and the amount of calculation in the case of executing the filter processing without dividing the image data,
An image processing method characterized by causing each process to execute a filter process based on an evaluation result. On the computer,
When a processor that performs a vector operation on image data including a plurality of elements executes filter processing on image data that includes fewer elements than the plurality of elements,
Based on the size of the image data including the elements smaller than the plurality of elements and the filter size applied in the filtering process, the filtering process is executed in parallel on each divided image data obtained by dividing the image data. Calculating the amount of calculation for each of the plurality of division methods, and evaluating the calculated amount of calculation and the amount of calculation in the case of executing the filter processing without dividing the image data,
An image processing program for executing each process for executing a filter process based on an evaluation result.

Images