JPH08172629A - Moving vector detector - Google Patents

Abstract

PURPOSE: To reduce the power consumption and to decrease a waiting time for reading out data by sequentially replacing data calculated and not referenced any more with data required for an image block processed next. CONSTITUTION: A retrieval buffer 14 reads out picture element data for all retrieval range with respect to a 1st micro block MBI prior to processing of the MBI. A control circuit 18 reads out a first element of the MBI and makes arithmetic operation with a prescribed part of the picture element data within the retrieval rage of the MBI. Furthermore, a processor 15 makes arithmetic operation of a difference absolute value between the initial picture element data of the MBI and all data received by the circuit 18. Then elements of the MBI are shifted downward by one row and calculated and the elements are moved to those to the right and stored in an accumulator. Then the arithmetic operation result is obtained as a sum of all differences. The data already calculated and not referenced any more are replaced with data required by an image block processed next.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to coding of a digitized moving picture signal, and more particularly to a motion vector detecting device for efficiently detecting a motion vector.

[0002]

2. Description of the Related Art Digitized moving images have a great deal of information. Therefore, conventionally, a method has been adopted in which a moving image signal is highly efficiently encoded and compressed. One of them is inter-picture predictive coding that predicts a current picture from previous or subsequent pictures by utilizing the correlation between pictures of picture signals.
At this time, it is the motion vector that is used to obtain the optimum predicted image. The motion vector is a displacement on the screen between the pixel block data being encoded and the predicted image data, and the sum of absolute values of the difference values between the original image and the reference image or the sum of squares of the difference values is used as the evaluation value, and the search is performed. The motion vector of the pixel block (macro block) having the smallest evaluation value within the range is set.

FIG. 6 shows a method of obtaining a motion vector according to a conventional technique by a full search method. Between pixel data in the search range for all vectors of a macroblock,
A predetermined operation is performed on the vector (x, y) and the pixel data of the macro block. And the processing here repeats the calculation for the number of possible vectors. As shown in FIG. 6A, the processing order is evaluated from the upper left vector, and then the vector is changed downward and sequentially evaluated (FIG. 6B), and the lower left vector is evaluated. All the evaluations in the first column are complete (FIG. 6 (c)). Next, FIG.
As shown in (d), the vector shifted to the right by one column in FIG. 6 is evaluated. After that, the vector to be evaluated is repeatedly moved from top to bottom and rightward to evaluate all the vectors.

[0004]

In the motion vector detection by the full search method without taking any measures, when the processing of one macroblock is finished and the processing of the next macroblock is started, the search range of the macroblock is changed. It is necessary to newly read pixel data. However, the search range for a certain macroblock generally overlaps with the search ranges for adjacent macroblocks. However, in the conventional method, it is necessary to read the same data a plurality of times from the external memory via the bus, and this is particularly wasteful in terms of power consumption. Further, since it is necessary to read data every time, there is a problem that the operation speed does not increase due to this data reading operation.

A method of simultaneously processing two macroblocks has been proposed, paying attention to the fact that the data referred to by adjacent macroblocks overlap. Since the motion vector detecting device according to this method is configured to use the same data for two macro blocks, the data read amount is halved when the target macro block is processed in the vertical direction. . However, there is a problem in that when the columns change, it is necessary to newly replace all the data in the search range.

In view of these problems, the present invention provides an apparatus capable of reducing the power consumption of the system and the waiting time for reading data by efficiently reading the pixel data in the search range. Is intended.

[0007]

In order to achieve the above object, the motion vector detection device of the present invention reads out pixel data of an exploration range required by an arbitrary pixel block and processes the pixel block. At the same time, of the data of the pixel block, data which has already been calculated and is not referred to in the future is sequentially replaced with data required by the pixel block to be processed next.
Further, in the motion vector detecting device, the reading of each pixel of the pixel block and the processing of the pixel block are
The feature is that it is performed in a Miyanda type.

[0008]

According to the present invention, since the pixel data in the search range referred to by the adjacent macroblocks are mostly overlapped, all the data are exchanged when processing the neighboring macroblocks. No need. Further, the pixels in the macroblock are processed pixel by pixel, and the macroblocks are processed in the order of the meander type. Therefore, the amount of overlapping data is always the same, unnecessary data transfer can be avoided, and power consumption can be reduced. In addition, it is necessary to read all the data from the outside during the operation for the first macroblock in one screen, but when performing the operation for macroblocks in the vicinity, the time required for data loading is apparently 0. Since the data is read so that, the waiting time for data loading can be shortened when the processing of the next macroblock is started.

[0009]

EXAMPLES The present invention will be described in detail below with reference to examples.

FIG. 1 shows the configuration of a motion vector detecting device according to the present invention. In FIG. 1, 11 is a picture memory that stores the current image as digitized data, and 12 is a picture memory that stores the image referred to by the current image. Since both 11 and 12 store an image for one screen, a large-capacity memory such as DRAM is generally connected to the outside via a bus or the like. Reference numeral 13 is a buffer memory that stores data for one pixel, and 14 is a buffer memory that stores pixel data in a search range for a certain macroblock. The size of the macroblock is 16 × 1.
6. When the search range is ± i, it has a capacity for storing the entire search range referred to by the macroblock, that is, (2i + 16) ² pixel data. The buffers 13 and 14 are READ / WRITE controlled by the control circuit 18. The processor 15 has a predetermined calculation and accumulation function such as a sum of absolute differences or a sum of squared differences between the pixel data of the macroblock and the pixel data of the search range. The minimum value detection circuit 16 is a circuit for calculating the minimum value from the (2i + 1) ² data calculated by the processor 15, and is configured to calculate the motion vector for the macroblock from the result. Hereinafter, the number of macroblocks in one screen is m × n.

Next, each operation in the configuration of FIG. 1 will be specifically described.

First, the first macroblock (hereinafter referred to as MB1
Process) will be described. Pixel data in the entire search range for MB1 is read from the picture memory 12 into the search buffer 14 before the processing of MB1 is started. The number of data is the number of pixel data in the entire search range, that is, (2i + 16) ² pixels. In the following, the right direction is the x-axis direction and the downward direction is the y-axis direction, and the element (x, y) represents the element number of the macroblock being processed.

The first element (0,0) of MB1 is read out by the control circuit 18 and is operated with the pixel data in the search range. Of the entire search range of MB1, the pixel data on which the element (0,0) is actually calculated is the shaded area in FIG. 2A, and the number is (2i + 1) ² . In the processor 15, the element (0,0) of MB1
Pixel data of the processor 15 and the control circuit 18
A predetermined calculation such as the absolute difference value is performed with all the data in the shaded area in FIG. 2 (a) input to the above, and the calculation result is stored in an internal accumulator or the like. The number is MB
It is equal to the number of pixel data (2i + 1) ² in the search range calculated for the element (0, 0) of 1.

Next, regarding the element (0, 1) of MB1,
The same calculation is performed with the pixel data in the search range. The pixel data of the search range calculated as the element (0, 1) of MB1 is the range shown by the shaded area in FIG. 2B, and the pixel data of the search range calculated by the element (0, 0) is 1 It is shifted down by a line. After that, the addition result is calculated with the operation result of the previous element stored in the accumulator,
Write it in the accumulator newly.

Similarly, elements (0, 2) and (0, 3)
The calculation is performed while sequentially shifting the elements of MB1 downward along the y-axis. As shown in FIG. 2 (c), when the maximum value in the y-axis direction, that is, the element (0, 15) of MB1 has been calculated, next, as shown in FIG. 1, 15) is calculated. The pixel data of this search range is obtained by shifting the search range of the element (0,15) by one column to the right. After that, by the control circuit 18, FIG.
When the calculation is performed up to the last element (15,0) in the order of (a), a total of (2i + 1) ² calculation results are obtained as the sum of the difference values for all the vectors. Then, finally, the minimum value detection circuit 16 detects the minimum value from the (2i + 1) ² pieces to obtain a motion vector.

Then, the next macroblock (hereinafter referred to as MB2)
And the pixel data in the search range. FIG. 4 shows pixel data of the entire search range referred to by MB1 and MB2. The solid line part 403 is the pixel data of the entire search range of MB1, and the broken line part 404 is the pixel data of the full search range of MB2. As shown in the figure, the pixel data of the entire search range for MB2 largely overlaps with the pixel data of MB1. As described above, in the method of performing the calculation for each pixel according to the order of FIG. 3A, when the element to be calculated changes in the x-axis direction, the buffer memory 14
The pixel data in the leftmost column, that is, the pixel data in the x-i-th column of the entire search range of MB1 is not referred to by an element whose x axis of MB1 is 1 or more, or all elements of MB2. In other words, the pixel data in the -i column of the search range is referred to by the element (0, a) of MB1 (where a is 0).
~ 15) only. Therefore, the elements of MB1 (0,
It can be rewritten during the calculation of the second column, that is, the calculation of the element (1, a) (a is 0 to 15) after the calculation of 15) is completed. The pixel data of the search range to be newly written in the buffer 14 is the pixel data which is not referred to by MB1 but is referred by MB2, that is, the data after the 2i + 16th column of the pixel data of the search range of MB1. In other words, since the area 401 is no longer needed when the processing of the element (0, 15) of MB1 is completed, the pixel data of the area 402 is newly added to the portion of the pixel data written in 401 by the control circuit 18. Perform the write operation.
After that, the control circuit 18 performs the rewriting operation when the element to be calculated in the x-axis direction of MB1 is deviated. MB
When the process of 1 ends and the process of MB2 starts, the contents of the buffer 14 are replaced with the pixel data of the search range of MB2.

When performing a rewriting operation according to the above method, the number of pixel data read from the external picture memory 12 into the search buffer 14 when the macroblock to be processed changes is only (2i + 16) × 16. Considering further one screen, (2i + 16) ² pixels are required when the macro block to be processed first is calculated, so a total of (2i + 16) ² + (m × n−
Data for 1) × {16 × (2i + 16)} pixels is read. With the simplest method, the number of data transfers when reading data in the entire search range each time is (2i + 16) ² ×
Since m × n, m × n = 45 × 30, the search range is ±
When set to 16, about 30%, when set to ± 32, about 2
The amount of data to be read can be reduced. Even in the conventional method of simultaneously processing two macroblocks, the amount of data to be read can only be reduced by half. Therefore, the number of times the bus is driven is reduced accordingly, and power consumption can be reduced. Further, when the processing of MB2 is started, the data in the search buffer 14 has already been replaced with the data in the search range of MB2.
There is no time lag between the two processes and the processing speed does not decrease.

When processing all the macroblocks of one screen, the order of macroblock processing is performed in the form of a mounder as shown in FIG. 5 (a). The method described above is
2 is at the right position in the x-axis direction with respect to MB1, but when it is at the left position, the calculation is performed in the reverse order as shown in FIG. Also, the macroblock at the position (k, m-1) (where k is an even number) on the screen is shown in FIG. 3C, and the macroblock at the position (l, 0) (where l is an odd number) is shown in the figure. The calculation of each element is performed in the order of 3 (d).

The buffer 14 is preferably a dual port memory capable of reading and writing simultaneously. Further, in the present embodiment, since the calculation between one pixel and the pixels in the plurality of search ranges is performed, the processor 15 broadcasts the data for one pixel of the macroblock and can perform the calculation in parallel. Type parallel processor is preferred.

In this embodiment, the processing order of the macroblocks is described as FIG. 5A, but the motion vector is detected by the same operation even in the order in which the macroblocks are continuous as shown in FIG. 5B. be able to.

[0021]

According to the motion vector detecting apparatus of the present invention, the amount of pixel data read from an external picture memory can be significantly reduced, so that the power consumption of the system can be reduced. At the same time, when the processing is moved to the next macroblock, the necessary data has already been read, so the waiting time for data loading can be shortened and the speed can be increased.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a motion vector detection device according to the present invention.

FIG. 2 is a diagram showing pixel data in a search range referred to by each pixel of a macroblock.

FIG. 3 is a diagram showing an order of processing of pixels of a macro block.

FIG. 4 is a diagram showing pixel data in a search range referred to by two adjacent macroblocks.

FIG. 5 is a diagram showing an order of processing macroblocks.

FIG. 6 is a diagram showing a method of detecting a motion vector according to a conventional example.

[Explanation of symbols]

 11, 12 picture memory 13 buffer for storing one pixel of macroblock 14 buffer for storing pixel data in search range 15 calculator 16 minimum value detection circuit 18 control circuit 21, 22 bus

Claims (2)

[Claims] 1. A motion vector detecting device for detecting a motion vector of a pixel block in a digitized moving image, reads out pixel data of a search range required by an arbitrary pixel block, and processes the pixel block. At the same time, the motion vector is characterized in that, among the data of the pixel block, data that has already been calculated and is not referenced in the future is sequentially replaced with data required by the pixel block to be processed next. Detection device. 2. The motion vector detection device according to claim 1, wherein the pixel-by-pixel reading in the pixel block processing and the pixel block processing are performed in a meander type.