JPH0993587A - Motion vector retrieval method and retrieval device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain high speed vector retrieval processing by returning picture element data stored in a register among picture element data of a search window used for calculation of a distortion corresponding to one current image block to a processor element gain so as to calculate the distortion corresponding to the different current image block. SOLUTION: Distortion is calculated, based on picture element data of a search window and picture element data of a current image block. Simultaneously in the case of calculating distortion, part of picture element data in the picture element data of the search window to be transferred by a window data transfer unit 3000 are latched by a horizontal side register unit 5000. The picture element data of the search window to be latched are returned again to the window data transfer unit 3000 and the distortion is calculated by the processor unit, based on the search window picture element data and the picture element data of the current image.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion vector search device applied to information compression of digital moving images,
In particular, a motion vector for searching a motion vector by each distortion calculated based on pixel data of a current image block forming a part of the current image and pixel data of a plurality of candidate blocks in a search window on the reference image. The present invention relates to a search device.

[0002]

2. Description of the Related Art In recent years, multimedia which expresses different information such as characters, figures, voices, and images by digital data, and integrates these media and handles them in a unified manner has attracted attention in recent years. One of the key technologies to realize this multimedia more effectively is information compression technology.
Information compression technology is a technology that focuses on information redundancy and reduces the amount of information by reducing the information in redundant parts, thereby efficiently processing, storing, and transmitting a large amount of information. It becomes possible.

[0003] There is a large difference in the amount of information of various media. In particular, since a moving image has a huge amount of information, a large amount of information compression is required. There are various methods for compressing information, and by combining these methods,
Significant compression has been achieved. Generally, these compression functions are provided as an LSI (Large scale integrated circuit).

As one of the compression methods for moving image information, there is known a method of compressing information based on a temporal correlation between two screens constituting a part of a moving image. This compression method includes, for example, a simple inter-frame predictive coding method and a motion-compensated inter-frame predictive coding method. Here, a frame represents one screen constituting a moving image. Less than,
The simple inter-frame prediction coding method and the motion compensation inter-frame prediction coding method will be described.

FIG. 39 is a diagram showing a simple interframe predictive coding system. In the simple inter-frame predictive coding method, a difference value between each pixel data of the current image 100 and each pixel data of the reference image 300, which positionally correspond to each other, is calculated, and this difference value is compared with an appropriate threshold value. , Significant pixel data and insignificant pixel data. The significant pixel data corresponds to the case where this difference value is larger than the threshold value, and is data that cannot be omitted when predicting the current image 100 from the reference image 300. On the other hand, insignificant pixel data is
This difference value is equal to or less than the threshold value, and the reference image 300
It is data that can be reduced when the current image 100 is predicted from.

The reference image 300 may be a past image or a future image relative to the present image 100, but is an image encoded earlier than the present image 100 in time. . For example, as shown in FIG. 39, when the person image 10 in the previous image 100 is moving to the right in the current image 100, the region showing the significant pixel data is shown by two significant pixel regions 20 and 21. Pixel data on the current image 100, which positionally corresponds to the significant pixel region 20, can be represented by the difference value between the pixel data and the significant pixel region 20 and the significant pixel region 20, and the significant pixel region 21 is positioned in the position. The corresponding pixel data on the current image 100 can be represented by the difference value between this pixel data and the significant pixel region 21 and the significant pixel region 21. The remaining non-significant pixel area can be represented by the pixel data itself of the reference image 300 that positionally corresponds to this non-significant pixel area.

In the simple inter-frame predictive coding method, the smaller the number of significant pixels, the smaller the data amount of the difference value required for prediction, so that the compression efficiency can be improved. Also, by setting a high threshold, the number of significant pixels can be reduced to further improve the compression efficiency.However, if the threshold is set too high, the image becomes jerky and unnatural, May appear as an afterimage, causing a problem that the image quality is significantly deteriorated.

As described above, in the simple inter-frame predictive coding method, the current image 100 is predicted based on the pixel data at the same position of the reference image 300, so that the image between the current image 100 and the reference image 300 is predicted. Although a high compression efficiency can be realized when the change is small, as shown in FIG. 39, when a part of the image moves significantly on the image, the compression rate is higher than that of the simple interframe predictive coding method. The compression efficiency is higher when the motion-compensated inter-frame predictive coding method described in 1. is used.

In the motion compensation interframe predictive coding system,
If the person image 10 moves as shown in FIG.
The motion vector MV indicated by 0 is calculated. The motion vector MV represents the moving direction and the moving distance of the person image 10, and the motion vector MV and the person image 1 of the reference image 300.
The person image 10 on the current image 100 is predicted by the pixel data forming 0. In this case, the significant pixel area is only the area 20. Therefore, the motion-compensated inter-frame predictive coding method can significantly reduce the number of significant pixels, and thus can significantly improve the compression efficiency of image information.

Incidentally, the ITU-T which is an international standard system
(International telecommunication Union-Telecommun
ication Standardization Sector) H. In the motion compensation inter-frame prediction method according to H.261, first, as shown in FIG. A search window 410 including a plurality of blocks 302 (hereinafter referred to as candidate blocks) is specified on the reference image 302, and the plurality of candidate blocks 302 and the current image block 22 included in the search window 410 are identified.
Calculate the distortion with 0.

Here, the distortion represents the similarity between each candidate block 302 and the current image block 220, and the difference value of the pixel data corresponding to each position in each candidate block is obtained, respectively. The difference value is shown as a value accumulated by being converted into positive number data by an absolute value calculation or a square calculation so as not to be offset. Next, the distortion having the minimum value is specified from the calculated distortions, and the motion vector MV is calculated based on the candidate block 302 having the minimum distortion and the current image block 220.

Further, the relationship among the current image block 220, the search window 410 and the candidate block 302 will be described. As shown in FIG. 42B, the current image block 2
If 20 is composed of pixels in N rows and M columns and the search window 410 is composed of pixels in H rows and L columns as shown in FIG. 42A, the candidate block 302 similar to the current image block 220 is , (H−N + 1) × (L−M + 1) pieces exist in the search window 410.

If the pixel data at the upper left corner of the current image block 220 is represented by a (0,0), each candidate block positionally corresponding to this pixel data a (0,0) in the search window 410. The range that can be taken by the pixel 302 is indicated by the hatched area in FIG. FIG. 43 shows the positional correspondence between the pixel data in the current image block 220 and the pixel data in each candidate block 302. FIG.
Pixel data a in the current image block 220
Pixel data in each candidate block 302 corresponding to (m, n) in position is represented by pixel data b (l + m, h + n) in the search window 410. Here, h and l are each candidate block 302 in the search window 410.
The pixel data b (l, h) in the search window 410 is the pixel data at the upper left corner of the candidate block 302, and is the pixel data a (0,0) at the upper left corner of the current image block 220. Positionally correspond.

In the current image block 220, the search window 410 and the plurality of candidate blocks 302 shown in FIGS. 42 and 43, the distortion between the current image block 220 and each candidate block 302 is D (l, h).
Then, D (l, h) is represented by the following equation.

[0015]

[Equation 1]

Here, ‖‖ represents the norm for calculating the distortion, and d (m, n) is represented by d (m, n) = b (l + m, h + n) -a (m, n), and The local distortion which is the difference value of the pixel data of the image block 220 and the pixel data of each candidate block 302 corresponding in position is shown. The norm operation generally uses an absolute value operation and a square operation, but the absolute value operation is most frequently used in terms of computational complexity and efficiency.

In the motion-compensated inter-frame prediction method, the method of comparing the current image and the reference image in block units is called a block matching method.
When all the candidate blocks included in the search window are compared with the current image block, it is called a full search block matching method (all-point search method).
As a method and an apparatus for obtaining a motion vector using the all-points search method, for example, a method and a circuit for processing a data signal representing a continuous image of a two-dimensional animated image disclosed in Japanese Patent Application Laid-Open No. Hei 2-213291 are known. I have. In this method and circuit, in order to reduce the operation time, processor elements are arranged by the number of candidate blocks,
The distortion is obtained by switching the pixel data of the search window supplied to the processor element in the upward, downward and leftward directions as a whole to perform scanning.

That is, as shown in FIGS. 44 and 45, if l and h are represented by l = 0,1,2,3 h = 0,1,2,3, first, the shaded 9 The pixel data of the search window is input to each of the processor elements, and the pixel data a of the current image block is input.
In cycle 0 in which (0,0) is input, the local distortion | b (l, h) -a (0,0) | is calculated and stored in each processor element.

In the next cycle 1, each pixel data of the search window is moved upward as a whole, and the pixel data a (0, 1) of the current image block is inputted, so that the local distortion | b (l, h + 1) ) −a (0, 1) | is further added to the local distortion calculated in cycle 0 and stored.

Next, in cycle 2, each pixel data of the search window is moved to the left as a whole,
By inputting the pixel data a (1,1) of the current image block, the local distortion | b (l + 1, h + 1) -a (1,1) | is calculated, and the calculation result in cycle 1 It is added and stored.

Next, in cycle 3, while moving each pixel data of the search window as a whole,
By inputting the pixel data a (1,0) of the current image block, | b (l + 1, h) -a (1,0) | is calculated and further added to the operation result in cycle 2. And as a result, the distortion between each candidate block corresponding to the nine candidate blocks and the current image block is calculated.

Then, the motion vector is obtained based on the minimum distortion detected from these distortions. Generally, in the process of searching a motion vector, the distortion calculation and the access to the image memory occupy a large part, and a high-speed LS is required to perform a huge amount of processing.
I is required. On the other hand, the motion vector search device adopts a two-dimensional systolic array structure and
Arranging multiple processor elements in the I-chip,
By performing the parallel operation, the speedup of the motion vector search process is realized.

[0023]

However, in the conventional motion vector search device, the motion vector search device shown in FIGS.
The pixel data of the search window shown in Fig. 4 is used for calculating the distortion, and disappears when scanned leftward of the processor element in the leftmost column, and sets the search window for each of a plurality of current image blocks. When calculating the distortion by using the pixel data of the same search window, it is necessary to access the image memory in which the pixel data of the reference image is stored many times. There is a problem that it takes a longer time to access the pixel data of the window to the image memory, and the access speed of the pixel data of the reference image to the image memory hinders the improvement of the overall processing speed.

Therefore, according to the present invention, for a plurality of current image blocks sharing pixel data of the same search window, among the pixel data of the search window used for calculating the distortion corresponding to one current image block, By providing a register for holding pixel data that disappears from the processor element, returning the pixel data held in the register to the processor element again, and calculating the distortion corresponding to a different current image block, the pixel data of the search window is displayed. An object of the present invention is to provide a motion vector search method and a search device capable of reducing the amount of data accessing the memory and relatively speeding up the motion vector search process.

[0025]

In order to solve the above-mentioned problems, the present invention according to claim 1 is based on a current image partially forming a moving image based on a reference image partially forming the moving image. A method of searching a motion vector used for prediction, wherein the current image includes a plurality of current image blocks represented by pixels in a plurality of rows and a plurality of columns having pixel data, and the reference image has pixel data. The current image block includes a plurality of candidate blocks having the same size, and for each of the current image blocks, the block position of any one of the plurality of candidate blocks that is similar to the current image block and the current image block. A motion vector search method for specifying the motion vector according to a block position, the pixel data of the plurality of current image blocks, Preparing a plurality of search window pixel data including at least two candidate blocks among a plurality of candidate blocks of the reference image; and forming a search area according to the sizes of the search window and the current image block, First pixel data transfer holding means for inputting and holding a part of pixel data of each candidate block in the search window in the search area and transferring the held pixel data in a predetermined direction of the search area, Pixel data is transferred to and from the first pixel data transfer holding means, and the pixel data of the search window is transferred along with the first pixel data transfer holding means along a predetermined transfer path passing through the search area. Two pixel data transfer holding means, and a part of the pixel data transferred along the transfer path to the first and second pixels. A third pixel data transfer holding means for inputting and outputting from the data transfer holding means, and inputting pixel data of the search window to the first and second pixel data transfer holding means. , The pixel data of the search window is moved by the first and second pixel data transfer holding means so that the pixel data of each pixel column of the search window moves in the row direction while reciprocating in all columns together in the column direction. To the current image block based on the pixel data of the current image block and the pixel data held in the first pixel data transfer / holding means. On the other hand, each of a plurality of distortions representing the image difference between the current image block and the plurality of candidate blocks is calculated. A first distortion calculating step; and a similar block specifying step of detecting the minimum value of a plurality of distortion values calculated for each of the current image blocks and specifying the similar one candidate block. Including,
Further, among the pixel data of the search window input to the first and second pixel data transfer holding means, the pixel data of each column by a predetermined column from the first column of the search window is converted into the first and second pixel data. Discharge data holding step of causing the third pixel data transfer holding means to hold the discharged pixel data, and the pixel data held by the third pixel data transfer holding means. The pixel data returning step of returning from the transfer holding means to the first and second pixel data transfer holding means, and the pixel data returned to the first and second pixel data transfer holding means by the pixel data returning step are used. And a second distortion calculation step of performing distortion calculation corresponding to a current image block different from the current image block. The features.

According to a second aspect of the present invention, there is provided an apparatus for searching a motion vector used for predicting a current image partially forming a moving image based on a reference image partially forming the moving image. Yes, the current image includes a plurality of current image blocks represented by pixels in a plurality of rows and a plurality of columns having pixel data, and the reference image includes a plurality of candidate blocks having the same size as the current image block each having pixel data. , A motion vector that specifies the motion vector for each of the current image blocks by the block position of any one candidate block similar to the current image block among the plurality of candidate blocks and the block position of the current image block A search device, comprising a plurality of searches, each of which stores pixel data of the reference image and includes at least two candidate blocks. Reference image data storage means capable of outputting window pixel data, and a search area corresponding to the size of the search window and the current image block, and pixel data of each candidate block in the search window in the search area Between a first pixel data transfer holding means for inputting and holding a part of the stored pixel data and transferring the held pixel data in a predetermined direction of the search area, and the first pixel data transfer holding means. Exchange data,
A second pixel data transfer / holding means for transferring the pixel data of the search window together with the first pixel data transfer / holding means along a predetermined transfer path passing through the search area; and a second pixel data transfer / holding means for transferring the pixel data in the search window along the transfer path. Third pixel data transfer holding means for inputting and outputting a part of the pixel data from the first and second pixel data transfer holding means, and pixel data of the search window for the first and second pixel data transfer Inputting to the holding means, at least two candidate blocks of the plurality are sequentially input into the search area, and pixel data of each pixel column of the search window is reciprocally moved in all columns together in the row direction. So that the pixel data of the search window is moved along the transfer path by the first and second pixel data transfer holding means. A window data transfer control means for creating transmission, wherein the pixel data first current picture block
A plurality of distortions representing image differences between the current image block and the plurality of candidate blocks, based on the pixel data held in the pixel data transfer holding means of FIG. Distortion calculation means for calculating, a similar block specifying means for detecting a minimum value among a plurality of distortion values calculated for each of the current image blocks, and specifying the similar one candidate block; When the pixel data is transferred by the first and second pixel data transfer holding means, one of the search windows among the pixel data input to the first and second pixel data transfer holding means
Ejection data holding control means for causing the third pixel data transfer holding means to hold the pixel data of each row for a predetermined number of rows from the first row while being discharged from the first and second pixel data transfer holding means, and the discharging. Return data transfer control means for returning the pixel data held in the third pixel data transfer holding means by the data holding control means from the third pixel data transfer holding means to the first and second pixel data transfer means And the distortion calculating means uses the pixel data returned from the third pixel data transfer holding means to the first and second pixel data transfer holding means by the return data transfer control means, It is characterized in that distortion calculation is performed for a current image block different from the current image block.

According to a third aspect of the present invention, in the second aspect of the invention, after the distortion calculation is completed by the distortion calculation means, the pixel data in the search window is returned so as to return to the state when the distortion calculation was started. The data transfer control means returns the pixel data of the search window held by the third pixel data transfer holding means to the first and second pixel data transfer holding means, and the window data transfer control means operates by the The pixel data of the search window held by the first and second pixel data transfer holding means is transferred in the direction opposite to the transfer path, and the distortion calculating means corresponds to the current image block different from the current image block. It is characterized in that distortion calculation is performed.

According to a fourth aspect of the present invention, there is provided a device for searching a motion vector used for predicting a current image partially forming a moving image based on a reference image partially forming the moving image. Yes, where N and M are integers, the current image includes a plurality of current image blocks represented by N rows and M columns of pixels having pixel data, and the reference image has each current image block having pixel data. And a block position of any one of the plurality of candidate blocks that is similar to the current image block and a block position of the current image block. By
A motion vector search device for specifying the motion vector, wherein when H and L are respectively integers, a plurality of H rows and L columns, each of which stores pixel data of the reference image and includes at least two candidate blocks, respectively. Reference image data storage means capable of outputting pixel data of a search window and a search area of (H-N + 1) rows (L-M + 1) columns corresponding to the sizes of the search window and the current image block are formed, and the search is performed. Part of the pixel data of each candidate block in the search window is input and held in the area, and the held pixel data is transferred in the predetermined direction of the search area (H−N + 1) × (L−M + 1). Pixel data is transferred between the first pixel data transfer holding means having a plurality of first registers and the first register of the first pixel data transfer holding means. Second pixel data transfer holding means for transferring the pixel data of the search window along a predetermined transfer path passing through the search area together with the first pixel data transfer holding means. , A third portion having (M-1) × H third registers for inputting and outputting a part of the pixel data transferred along the transfer path from the first and second pixel data transfer holding means. Pixel data transfer / holding means and pixel data of the search window are input to the first and second pixel data transfer / holding means, and at least two candidate blocks of the plurality are sequentially input into the search area, and In order to move the pixel data of each pixel column of the search window in the row direction while reciprocating in all columns together, the first and second pixel data transfer protections are performed. Based on the window data transfer control means for transferring the pixel data of the search window along the transfer path by means, and the pixel data of the current image block and the pixel data held in the first pixel data transfer holding means. Then, for each of the current image blocks, distortion calculation means for respectively calculating a plurality of distortions that represent image differences between the current image block and the plurality of candidate blocks, and for each of the current image blocks. Pixel data is transferred by a similar block specifying unit that detects the minimum value of the calculated plurality of distortion values and specifies one similar candidate block, and the first and second pixel data transfer holding units. Of the pixel data input to the first and second pixel data transfer holding means. The pixel data of each column from the first column to the (M-1) th column of the search window is held in the third pixel data transfer holding unit while being discharged from the first and second pixel data transfer holding unit. The discharge data holding control means, and the pixel data held in the third pixel data transfer holding means by the discharge data holding control means from the third pixel data transfer holding means to the first and second pixel data transfers. Return data transfer control means for returning to the means, wherein the distortion calculation means returns from the third pixel data transfer holding means to the first and second pixel data transfer holding means by the return data transfer control means. Distortion calculation corresponding to a current image block different from the current image block is performed using the generated pixel data.

According to a fifth aspect of the invention, in the invention according to the fourth aspect, the return data transfer control means causes the pixel data for the (M-1) th column to be transferred from the third register to the first and second pixels. At the same time as being returned to the register, the pixel data retained in the first and second registers corresponding to the first pixel column of the search area is transferred so as to be returned to the other side in the row direction, and further, these M When the pixel data for the column is returned to the position at the start of the distortion calculation corresponding to any of the current image blocks, the transfer direction of the input pixel data is switched again to the one side in the row direction, and the distortion calculation means , (M-1) column pixel data returned from the third register to the first and second registers and the (M-1) column pixel data are input to the third register. Then, the distortion calculation corresponding to the current image block different from the current image block is performed using the pixel data held in the first and second registers corresponding to the first pixel column of the search area. It is characterized by starting.

According to a sixth aspect of the present invention, there is provided an apparatus for searching a motion vector used for predicting a current image partially forming a moving image based on a reference image partially forming the moving image. Yes, where N and M are integers, the current image includes a plurality of current image blocks represented by N rows and M columns of pixels having pixel data, and the reference image has each current image block having pixel data. And a block position of any one of the plurality of candidate blocks that is similar to the current image block and a block position of the current image block. By
A motion vector search device for specifying the motion vector, the reference capable of storing pixel data of the reference image and outputting pixel data of a plurality of H rows and L columns of search windows each including at least two candidate blocks. Image data storage means and (H-N + 1) rows (L-M) corresponding to the sizes of the search window and the current image block.
+1) forming a search area of columns, inputting and holding a part of pixel data of each candidate block in the search window in the search area, and transferring the held pixel data in a predetermined direction of the search area (H-N + 1) x (L-
M + 1) first pixel data transfer holding means having first registers, and a plurality of second registers for transferring pixel data between the first registers of the first pixel data transfer holding means, A second pixel data transfer / holding means for transferring the pixel data of the search window together with the first pixel data transfer / holding means along a predetermined transfer path passing through the search area; The third pixel data transfer holding means having M × H third registers for inputting and outputting a part of the pixel data from the first and second pixel data transfer holding means, and the pixel data of the search window While inputting to the first and second pixel data transfer holding means, at least two candidate blocks of the plurality are sequentially input into the search area. The pixel data of the search window is transferred by the first and second pixel data transfer holding means so that the pixel data of each pixel column of the search window moves in the row direction while reciprocating in the column direction together with all rows. Based on the window data transfer control means for transferring along the transfer path, and the pixel data of the current image block and the pixel data held by the first pixel data transfer holding means, for each of the current image blocks. , Distortion calculation means for respectively calculating a plurality of distortions representing image differences between the current image block and the plurality of candidate blocks, and a plurality of distortion values calculated for each of the current image blocks Similar block specifying means for detecting a minimum value and specifying one similar candidate block; 1st and 2nd
When the pixel data is transferred by the pixel data transfer holding means, the pixels of each column from the first column to the Mth column of the search window among the pixel data input to the first and second pixel data transfer holding means Ejection data retention control means for retaining data in the third pixel data transfer retention means while ejecting data from the first and second pixel data transfer retention means, and the third pixel data transfer retention means by the ejection data retention control means. Return data transfer control means for returning the pixel data held in the means from the third pixel data transfer holding means to the first and second pixel data transfer means,
The distortion calculation means uses the pixel data returned from the third pixel data transfer holding means to the first and second pixel data transfer holding means by the return data transfer control means, and It is characterized in that the distortion calculation corresponding to the current image block different from the image block is performed.

According to a seventh aspect of the present invention, in the invention according to the sixth aspect, the return data transfer control means sets pixel data for M columns from the first column to the Mth column of the search window to the first column. When returning from the 3 register to the position at the time of starting the distortion calculation corresponding to any one of the current image blocks, the transfer direction of the input pixel data is switched again to the one side in the row direction, and the distortion calculating unit It is characterized in that the distortion calculation corresponding to the current image block different from the current image block is started using the pixel data of M columns returned from the 3 register to the 1st and 2nd registers.

The invention described in claim 8 is an apparatus for searching a motion vector used for predicting a current image partially forming a moving image based on a reference image partially forming the moving image. Yes, the current image includes a plurality of current image blocks represented by pixels in a plurality of rows and a plurality of columns having pixel data, and the reference image includes a plurality of candidate blocks having the same size as the current image block each having pixel data. , A motion vector that specifies the motion vector for each of the current image blocks by the block position of any one candidate block similar to the current image block among the plurality of candidate blocks and the block position of the current image block A search device, comprising a plurality of searches, each of which stores pixel data of the reference image and includes at least two candidate blocks. Reference image data storage means capable of outputting window pixel data, and a search area corresponding to the size of the search window and the current image block, and pixel data of each candidate block in the search window in the search area Between a first pixel data transfer holding means for inputting and holding a part of the stored pixel data and transferring the held pixel data in a predetermined direction of the search area, and the first pixel data transfer holding means. Exchange data,
A second pixel data transfer / holding means for transferring the pixel data of the search window together with the first pixel data transfer / holding means along a predetermined transfer path passing through the search area; and a second pixel data transfer / holding means for transferring the pixel data in the search window along the transfer path. Third pixel data transfer / holding means for inputting and outputting a part of the pixel data from the first and second pixel data transfer / holding means, and one of the plurality of search windows from the reference image data storage means. Pixel data is read out and stored for each predetermined pixel area including a part of pixel data, and the stored pixel data of the search window is transferred at a transfer rate higher than the data transfer rate from the reference image data storage means. High-speed transfer storage means to be supplied to the second pixel data transfer holding means, and a search window stored in the high-speed transfer storage means Wherein the pixel data first and second
The pixel data transfer holding means for sequentially inputting at least two candidate blocks of the plurality into the search area, and reciprocally moving the data of each pixel column of the search window in the column direction together in all rows. While moving in the row direction, window data transfer control means for transferring the pixel data of the search window by the first and second pixel data transfer holding means along the transfer path, and pixel data of the current image block. Based on the pixel data held in the first pixel data transfer holding means, for each of the current image blocks, a plurality of image data representing a difference in image between the current image block and the plurality of candidate blocks. Distortion calculation means for calculating the distortion of each of the current image blocks, and By detecting the minimum value among the plurality of distortion values, and similar block specifying means for specifying a single candidate block the similar, wherein the first and second
When the pixel data is transferred by the pixel data transfer / holding means, among the pixel data input to the first and second pixel data transfer / holding means, the pixel data of each column by a predetermined number from the first column of the search window. Is stored in the third pixel data transfer holding means while being discharged from the first and second pixel data transfer holding means, and the third pixel data transfer holding means by the discharge data holding control means. Return data transfer control means for returning the pixel data held by the third pixel data transfer holding means to the first and second pixel data transfer means,
Pixel data in a transfer order following the pixel data returned in the search area among the pixel data for the predetermined pixel area read and stored from the reference image data storage means by the high-speed transfer storage means. The first and second
Pixels returned to the first and second pixel data transfer / holding means from the third pixel data transfer / holding means by the return data transfer control means. The distortion calculation corresponding to the current image block different from the current image block is performed using the data and the pixel data re-input from the high-speed transfer storage means to the first and second pixel data transfer holding means. It is characterized by

According to a ninth aspect of the invention, in the eighth aspect of the invention, the pixel data is returned from the third pixel data transfer holding means to the first and second pixel data transfer holding means, and Pixel data in a transfer order subsequent to the pixel data is re-input from the high-speed transfer storage means to the first and second pixel data transfer holding means, so that some of the pixels in one of the plurality of search windows When data is supplied to the search area, the remaining pixel data of any one of the search windows is read from the reference image data storage means and stored in the high-speed transfer storage means.

The invention according to claim 10 is the invention according to claims 4 to 9.
In any of the above aspects, the third pixel data transfer holding unit inputs and outputs the pixel data while maintaining the position of the pixel data at the start of the distortion calculation in the search area. The invention described in Item 11 is the invention described in any one of Items 2-5 and 8-10, wherein each of the current image block and the candidate block has an even number of pixel columns.

The invention described in claim 12 is the invention described in any one of claims 2 and 3 and 6 to 10, characterized in that the current image block and the candidate block each have an odd number of pixel columns. . According to a thirteenth aspect of the present invention, in the invention according to any one of the fourth to twelfth aspects, the distortion calculation means includes a systolic array having a plurality of arithmetic units arranged two-dimensionally together with the first register. It is characterized by being configured by an arithmetic circuit having a structure.

The invention described in claim 14 relates to claims 4 to 1.
3. In the invention according to any one of 3 above, (N−1) × (L−M + 1) pieces are provided so that some of the second registers have the same number in each column among the plurality of second registers, and the search area is provided. The pixel data transferred in one of the column directions is input and output, and (N−1) × (L−M + 1) pieces of remaining second registers are provided so that each column has the same number. The pixel data transferred in the other direction of the column is input and output.

The invention described in claim 15 is the invention according to claims 4 to 1.
3. In the invention according to any one of 3 above, the second registers are provided in (N-1) * (L-M + 1) pieces so that each column has the same number, and the second registers correspond to pixels at both ends in the column direction of the search area. It is characterized in that pixel data is transferred via one register.

[0038]

Preferred embodiments of the present invention will be described below with reference to the drawings. As shown in FIG. 1, the motion vector search device of the present embodiment forms a search area corresponding to the size of the search window and the current image block, and the pixel data of each candidate block in the search window is formed in this search area. While inputting and holding a part,
Pixel data is transferred between the first pixel data transfer holding means 1 for transferring the held pixel data in the predetermined direction of the search area, and the first pixel data transfer holding means 1, and the first pixel data transfer holding is performed. Second pixel data transfer holding means 2 for transferring the pixel data of the search window along with the means 1 along a predetermined transfer path passing through the search area, and pixel data of the search window for the first pixel data transfer holding means 1 and No. 2 pixel data transfer / holding means 2 to sequentially input at least two candidate blocks out of the plurality into the search region, and the pixel data of each pixel column of the search window reciprocates in all columns together in the column direction. While moving in the row direction, the pixel data of the search window is moved by the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2. Each of the current image blocks is based on the window data transfer control means 4 for transferring along the sending path, the pixel data of the current image block, and the pixel data of the search window held in the first pixel data transfer holding means 1. On the other hand, of the distortion calculation means 5 for calculating a plurality of distortions representing the image difference between the current image block and the plurality of candidate blocks, and the plurality of distortion values calculated for each of the current image blocks. And a similar block specifying unit 6 that detects a minimum value and specifies one similar candidate block. The motion vector search device first obtains pixel data of a desired search window from the reference image memory (not shown) in which the pixel data of the reference image is stored by the window data transfer control unit 4 by the first pixel data transfer holding unit 1 and the first pixel data transfer holding unit 1. 2 pixel data transfer holding means 2 and then at least two candidate blocks among a plurality of them are sequentially input into the search area, and the pixel data of each pixel column of the search window are reciprocated in the column direction in all rows together. The pixel data of the search window is transferred along the transfer path by the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2 so as to move in the row direction while moving.

Next, the pixel data of the desired current image block is input from the current image memory (not shown) in which the pixel data of the current image is stored by the distortion calculation means 5, and the pixel data of the input current image block and the Based on the pixel data of the search window held in one pixel data transfer holding means 1, for each of the current image blocks, a plurality of distortions representing image differences between the current image block and the plurality of candidate blocks. Are calculated respectively.

Next, the similar block specifying means 6 detects the minimum value of the plurality of distortion values calculated for each of the current image blocks, and specifies one similar candidate block. The motion vector search device further inputs a part of the pixel data transferred along the transfer path from the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2 and outputs the third pixel data. Pixel data transfer holding means 3 and first pixel data transfer holding means 1
And when the pixel data is transferred by the second pixel data transfer holding means 2, the search window among the pixel data input to the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2 The first pixel data transfer holding means 1 and the second pixel data transfer holding means 2 for the pixel data of each row from the first row
The discharge data holding control means 7 that is held in the third pixel data transfer holding means 3 while being discharged from, and the pixel data held in the third pixel data transfer holding means 3 by the discharge data holding control means 7. Return data transfer control means 8 for returning from the third pixel data transfer holding means 3 to the first pixel data transfer means 1 and the second pixel data transfer means 2 is provided.

The motion vector searching apparatus calculates the distortion of the pixel data of the search window held in the third pixel data transfer holding means when the distortion is calculated by the distortion calculating means 5, and then again calculates the first pixel data transfer holding means 1. By returning to the second pixel data transfer holding means 2, the distortion calculation corresponding to the current image block different from the current image block for which the distortion was first calculated is performed.

Now, let H, L, N and M be integers, and H
When searching for a motion vector corresponding to the current image block composed of pixels of N rows and M columns from a search window composed of pixels of row L columns, the search range is (H−N + 1) × (L−M
It is represented by +1) candidate blocks. The distortion calculation means 5 has the same number of arithmetic units as these candidate blocks, and is configured so that the distortions corresponding to the respective candidate blocks and the current image block are simultaneously calculated by parallel processing in each arithmetic unit.

When it is desired to obtain the motion vector from a wider search range, the motion vector can be obtained from the wider search range by increasing the number of arithmetic units to configure the motion vector search device. To obtain a search range twice as large as this search range, the search range is divided into a left half search range and a right half search range. It is possible to prevent the circuit scale from becoming huge.

That is, as shown in FIG. 2, the search range 511 of the current image block 201 is divided into the search range 501 and the search range 502, and the current image block 202 adjacent to the horizontal scan side of the current image block 201 is divided. Search range 51
2 is divided into a search range 502 and a search range 503,
The search range 513 of the current image block 203 is the search range 5
By dividing into 03 and the search range, the motion vector corresponding to each current image block can be obtained.

In this case, each search range 501, 502, 5
Since the number of columns 03, 504 (L−M + 1) is equal to the number M of columns of the current image blocks 201, 202, 203, the search range 502 is the right half search range of the current image block 201 and the current image block 201. The search range of the left half of 202, the search range 503 is the search range of the right half of the current image block 202, and the search range of the left half of the current image block 203.

That is, each current image block shares the search window of the search range of either the right half or the left half with the adjacent current image blocks to calculate the distortion. The motion vector search device first calculates the distortion corresponding to the search range 501 and the current image block 201, and calculates the distortion corresponding to the search range 502 and the current image block 201. Next, the search range 502 and the current image block 2
The distortion corresponding to 02 is calculated, and the search range is 5
03 and the distortion corresponding to the current image block 202 are calculated. After that, the pixel data corresponding to each of the search range and the current image block is alternately used twice, and the distortion is sequentially calculated to obtain the motion vector corresponding to each current image block from the double search range. it can.

In this case, the number of columns in the search range (LM
Although the number of columns M of the current image block is equal to +1), the number of columns (L-M + 1) of the search range is preferably an integral multiple of the number M of columns of the current image block. Thus, for multiple current image blocks that share the same search window pixel data,
The distortion is calculated from the pixel data of the search window and the pixel data of one current image block, and the pixels of the search window discharged from the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2 at this time. The data is held in the third pixel data transfer / holding means 3, and the held pixel data in the search window is returned to the first pixel data transfer / holding means 1 and the second pixel data transfer / holding means 2 again to obtain a different current image. Since the distortion corresponding to the block can be calculated, the amount of data for accessing the pixel data of the desired search window to the reference image memory (not shown) can be reduced, and the access time of the pixel data of the search window can be relatively reduced. Can be shortened. Therefore, the motion vector search process can be speeded up.

Further, the motion vector search device reads and stores pixel data for each predetermined pixel area including a part of pixel data of a desired search window from a reference image memory (not shown) in which pixel data of a reference image is stored. At the same time, the stored pixel data of the search window is supplied to the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2 at a transfer rate higher than the data transfer rate from the reference image memory. It is desirable to include the transfer storage means 9. Since the pixel data of the search window is huge compared to the pixel data of the current image block, the high-speed transfer storage means 9 can efficiently access the pixel data.

That is, the pixel data of the desired search window is read from the reference image memory in advance and stored in the high-speed transfer storage means, and when the distortion corresponding to the current image block is calculated, the reference image memory is not accessed and the high-speed transfer is performed. By repeatedly accessing the storage means, the pixel data of the search window can be supplied to the first pixel data transfer holding means 1 and the second pixel data transfer holding means 2. Therefore, the pixel data of the search window can be transferred at high speed to the first and second pixel data transfer holding means. Further, since the reference image memory does not need to be a high speed memory, it can be configured with an inexpensive memory and the cost of the memory can be reduced.

[0050]

Embodiments of the present invention will be described below with reference to the drawings. (Embodiment 1) FIGS. 3 to 34 are diagrams showing a motion vector search device according to Embodiment 1 of the present invention.

As shown in FIG. 3, the motion vector search apparatus includes a current image memory 1000, a reference image memory 2000,
Window data transfer unit 3000, similar block specifying unit 4000, horizontal side register unit 5
000 and the system control unit 6000. The current image memory 1000 is composed of a volatile memory and stores the current image in frame or field units.

Here, as shown in FIG. 4, the current image is N
It is divided into a current image block consisting of pixels in rows and M columns, and coding is performed in block units. The current image block may have any size, but for ease of explanation, the current image block is a block composed of pixels in 3 rows and 3 columns. here,
The current image block is represented by horizontal scanning direction x, vertical scanning direction y, and coordinates (x, y). Each pixel data of the current image block 211 is represented by a (0,0), a (0 , 1), a (0,2), a (1,0), a (1,1), a (1,2), a (2,0), a (2,1), a (2, 2), each pixel data of the current image block 212 adjacent to the current image block 211 in the horizontal direction is represented by a (3,0), a (3,1), a (3,2), a (4 , 0), a (4,1), a (4,2), a (5,0), a (5,1), a (5,2). These pixel data are read by the system control unit 6000 through the output terminal R and transferred to the window data transfer unit 3000.

The reference image memory 2000 is composed of a volatile memory, and stores the reference image referred to by the current image to be encoded stored in the current image memory 1000, in frame or field units. Here, as shown in FIG. 4, the search window may be composed of pixels of arbitrary H rows and L columns as long as it is larger than the current image block, but for ease of explanation, it is composed of pixel data of 5 rows and 5 columns. Shall be configured. Further, each pixel data of the search window 401 has coordinates (x, y) similar to the current image block.
With b (0,0), b (0,1), b (0,2), b (0,3), b (0,4), b (1,0), b (1, 1), b (1,2), b (1,3), b (1,4), b (2,0), b (2,1), b (2,2), b (2,3 ), b (2,4), b (3,0), b (3,1), b (3,2), b (3,3), b (3,4), b (4,0) , b (4,1), b (4,2), b (4,3), b (4,4). These pixel data are read by the system control unit 6000 through the output terminals S0 and S1 of the reference image memory 2000 and transferred to the window data transfer unit 3000.

As a result, from the current image block 211 or 212 and the search window 401, (H-N + 1)
× (L−M + 1) = 9 candidate blocks 301 can be set. The window data transfer unit 3000 is
As shown in FIG. 5, the processor unit 310 is further added.
0, the first vertical side register unit 3200, the second vertical side register unit 3300, and the input register unit 3400.

The processor unit 3100 has 9 candidate blocks each corresponding to 9 candidate blocks in the search window.
Each of the processor elements PE, the pixel data of the candidate block and the pixel data of the current image block, which are positionally corresponding to each other, are input to each processor element PE, their difference values are calculated, and these difference values are corrected. The distortion is calculated by summing the number-transformed local distortions.

First vertical side register unit 3200
Each of the second vertical side register units 3300 is composed of six vertical side registers VS. Each vertical side register VS is a buffer provided so as to transfer the pixel data of each search window input to each processor element PE in the vertical direction in FIG. 5 as a whole.

The input register unit 3400 is composed of seven input registers IR, and the pixel data of the search window corresponding in position to the pixel data of the current image block is simultaneously applied to each processor element PE corresponding to each candidate block. It is a buffer provided to be input. Output terminal S of reference image memory 2000
The pixel data of the search window read through 0 and S1 are first input to the input register IR (3,0) and the input register IR (3,3), respectively, and then the other adjacent input registers IR, vertical. It is transferred to the side register VS or the processor element PE.

Here, each processor element PE, each vertical side register VS, each input register IR, and each horizontal side register HS described later have a vertical position in FIG. 5 with the processor element PE (0, 0) as the origin. x = −3, −2, −1,0,1,2,3, the horizontal position is y = −2, −1,0,1,2,3,4, and the coordinates (x, y) are It shall be attached.

That is, the window data transfer unit 3000 inputs the pixel data of the search window read through the output terminals S0 and S1 of the reference image memory 2000 into the input registers IR (3,0) and I, respectively.
The pixel data input to R (3, 3) are repeatedly transferred in the vertical direction and the left direction in FIG. 5 between the registers and the processor elements adjacent to each other and held therein. The distortion is calculated based on the pixel data of the candidate block and the pixel data of the current image block.

Further, the window data transfer unit 30
00 cooperates with each horizontal side register SH of the horizontal side register unit, which will be described later, so that the pixel data of the search window is also transferred to the right among the horizontal side registers HS, the vertical side registers VS, and the processor elements PE. It is repeatedly transferred and held. The similar block identifying unit 4000 inputs the distortion calculated by the processor unit 3100, and selects the minimum distortion M among the input distortions.
Processor element PE in which inDis is detected and the detected minimum distortion MinDis is calculated
The candidate block in the search window corresponding to the current image block is specified based on the position information of
Find V.

The horizontal side register unit 5000 is
The pixel data of the search window, which is composed of 15 horizontal side registers HS and is output from the processor unit 3100 and the second vertical side register unit 3300, is input, and the input pixel data is input between adjacent horizontal side registers HS. The pixel data held in each horizontal side register HS is repeatedly transferred to the right and repeatedly returned to the left in FIG. 5 and returned to the processor unit 3100 and the second vertical side register unit 3300.

The system control unit 6000 reads out desired pixel data from the current image memory 1000 and the reference image memory 2000 and transfers them to the window data transfer unit 3000, and at the same time, the window data transfer unit 3000 and the similar block specifying unit 4000.
And a signal to the horizontal side register unit 5000, respectively, to control the operation of each unit.

In the motion vector search device according to the first embodiment, as shown in FIG.
The distortion is calculated by the processor unit 3100 based on the pixel data of 1 and the pixel data of the current image block 211, and a part of the pixel data of the search window 401 transferred by the window transfer unit 3000 at the time of the distortion calculation. The data is held in the horizontal side register unit 5000, and then the motion vector MV is obtained by the similar block specifying unit 4000 based on the calculated distortion.

Next, the horizontal side register unit 500
The pixel data of the search window 401 held at 0 is returned to the search window transfer unit 3000 again,
The distortion is calculated by the processor unit 3100 based on the pixel data of the search window 401 and the pixel data of the current image 212, and then the motion vector MV is obtained by the similar block specifying unit 4000 based on the calculated distortion.

Next, the distortion is calculated by the processor unit 3100 based on the pixel data of the search window 402 shifted by M pixels in the horizontal scanning direction of the search window 401 and the pixel data of the current image 212, and then calculated. The motion vector MV is obtained by the similar block identifying unit 4000 based on the distortion.

After that, by returning the pixel data of the search window 401 held in the horizontal side register unit 5000 at the time of distortion calculation to the search window transfer unit 3000 again, one search window corresponds to two current image blocks. The distortion is calculated, and the motion vector MV is obtained by the similar block specifying unit 4000 based on the calculated distortion. In other words, the distortion is calculated from each of the two search windows for one current image block.

That is, the search window 401 is the search window corresponding to the current image block 212 and the search window corresponding to the current image 213, and the search window 402 is the current image block 21.
The search window corresponds to No. 2 and the search window corresponding to the current image 213. In other words, the current image block 212 is the search window 401.
Distortion is calculated based on the search window 402 and the search window 402, and the distortion of the current image block 213 is calculated based on the search window 402 and the search window 403.

Hereinafter, of the two search windows corresponding to the current image block, the search window located on the left side in FIG. 6 is referred to as a left half search window.
The search window located on the right side of is called the right half search window. Further, the distortion calculated from the left half search window of the current image block will be referred to as left half distortion, and the distortion calculated from the right half search window of the current image block will be referred to as right half distortion. In the current image block 211, only the right half distortion is obtained from the search window 401 which is the right half search window.

Next, the signals output from the system control unit 6000 to each unit will be described. As shown in FIG. 7, the system control unit 6000 has output terminals P1 to P10, and the window data transfer unit 3 receives the respective signals output through the output terminals.
000, the similar block specifying unit 4000, and the horizontal side register unit 5000.

These signals are binary pulse signals and represent 0 when they are low level and 1 when they are high level. Hereinafter, each signal will be described based on the time charts shown in FIGS. 8 to 12 show the search window transfer unit 3
000 and the horizontal side register unit 5000 are operation time charts, and R shown in these figures represents the current image memory 10 by the system control unit 6000.
00 represents the pixel data of the current image block read through the output terminal R of S00, and S0 and S1 are the reference image memory 200 by the system control unit 6000.
0 represents the pixel data of the search window read through the output terminals S0 and S1 of 0.

13 and 14 are time charts showing the operation of similar block specifying unit 4000. The clock pulse signal CK1 output through the output terminal P1 is the master signal of this device and has a period of 1
It has a pulse width of / 2. The operation of other signals is set with reference to this clock pulse signal CK1.

The system control unit 6000 reads desired pixel data from the current image memory 1000 and the reference image memory 2000 in synchronization with the down edge of the clock pulse signal CK1 and transfers it to the window data transfer unit 3000. The processor element PE and each register of the window data transfer unit 3000, and each register of the horizontal side register unit 5000 input and output the pixel data of the search window in synchronization with the rising edge of the clock pulse signal CK1.

Further, each processor element PE of the processor unit 3100 has a clock pulse signal CK1.
The pixel data of the current image block is input in synchronization with the rising edge of. As shown in FIG. 8, the pixel data of the first search window 401 read by the system control unit 6000 is the window data transfer unit 3
The pulses of the clock pulse signal CK1 input to 000 are counted as the first clock (period C1).

The pulse signal CK2 output through the output terminal P2 is a signal that performs the same operation as the clock pulse signal CK1. The pulse signal SU output through the output terminal P3 is supplied to each processor element PE of the window data transfer unit 3000 and each register VS, I.
Pixel data in the search window, which is output to R and is adjacent to each other, is controlled so as to be transferred between the respective processor elements PE and the registers VS and IR in the vertical direction as a whole. Here, the pulse signal S
When U is 0, the pixel data of the search window is transferred from top to bottom as a whole, and when the pulse signal SU is 1, the pixel data of the search window is transferred from bottom to top as a whole.

The pulse signal SU is the clock pulse signal C
It has a pulse width of 6 times the pulse width of K1 and rises in synchronization with the third clock's down edge of the clock pulse signal CK1, and rises up to the 18th clock with a cycle of 6 times the cycle of the clock pulse signal CK1. Is output to. Next, the clock pulse signal CK1 after two clocks
Rises in synchronization with the 20th clock's down edge,
In 18 clocks up to the 37th clock, it is output so as to rise sequentially in a cycle of 6 times. Then 2
It rises in synchronization with the down edge of the 39th clock of the clock pulse signal CK1 after the clock. That is, the pulse signal SU is the 1st pulse from the 20th clock to the 38th clock after the 20th clock of the clock pulse signal CK1.
Nine clocks are set as one cycle, and this cycle is sequentially repeated.

Similarly to the pulse signal SU, the other signals output through the output terminals P3 to P10 of the system control unit 6000, which will be described below, each cycle 19 clocks of the clock pulse signal CK1 after the 20th clock. , The same operation is sequentially repeated. The pulse signal SL output through the output terminal P4 is output to each processor element PE and each register VS, IR of the window data transfer unit 3000, and a search window is generated between each processor element PE and each register VS, IR adjacent to each other. The pixel data is controlled so as to be transferred to the left in FIG. 5 as a whole.

The pulse signal SL is the clock pulse signal C.
The pixel data of the first search window 401 having a pulse width twice that of K1 and input to the search window transfer unit 3000 is the system control unit 6.
It is read out by 000 and rises at the same time, then 18
It is output so that it rises up to the clock at a cycle three times as long as the cycle of the clock pulse signal CK1. Next, it rises in synchronization with the 20th clock's down edge of the clock pulse signal CK1 after 2 clocks, and the operation is repeated at a triple cycle in the 18th clock up to the 37th clock. Then, 39 of the clock pulse signal CK1 after two clocks
After the clock, the same operation is repeated from the 20th clock to the 38th clock.

The pulse signal SH output through the output terminal P5 is output to each horizontal side register HS of the horizontal side register unit 5000 and the processor element P in the first column of the window data transfer unit 3000.
The pixel data of the search window output from E (0, y) and the vertical side register (0, y) is repeatedly transferred to the left horizontal side register HS and held.

The pulse signal SH first rises in synchronization with the 10th clock down edge of the clock pulse signal CK1 and goes down in synchronization with the 11th clock down edge of the clock pulse signal CK1. Next, it rises in synchronization with the 15th clock down edge of the clock pulse signal CK1 and goes down in synchronization with the 17th clock down edge of the clock pulse signal CK1. Next, 3 from the 20th clock of the clock pulse signal CK1
In the 19th clock of the 8th clock, it rises at the 31st clock of the clock pulse signal CK1 and goes down at the 33rd clock of the clock pulse signal CK1. Then, after the 39th clock, the same operation as the 20th to 38th clocks is repeated.

The pulse signal SR output through the output terminal P6 is output to each processor element PE and each vertical side register VS of the window data transfer unit 3000, and each horizontal side register HS of the horizontal side register unit 5000, and mutually. Each adjacent processor element PE and each register HS, V
Pixel data in the search window between S as a whole is shown in FIG.
The control is performed so that the data is transferred in the right direction.
The pulse signal SR returns the pixel data of the search window held in each processor element and each register to the distortion calculation start state again after the distortion calculation.

The pulse signal SR first rises in synchronization with the 18th clock down edge of the clock pulse signal CK1 and goes down in synchronization with the 21st clock down edge of the clock pulse signal CK1. 19th from the 20th clock to the 38th clock of the clock pulse signal CK1
In the clock, 21 of the clock pulse signal CK1
After going down at the clock, clock pulse signal CK1
It rises at the 38th clock. Then, after the 39th clock, the same operation as the 20th to 38th clocks is repeated.

The pulse signal C output from the output terminal P7
L is output to each processor element PE of the processor unit 3100 and sets the initial state of the operation for calculating the distortion. Pulse signal CL
Has a pulse width twice the pulse width of the clock pulse signal CK1, and first rises in synchronization with the 9th clock down edge of the clock pulse signal CK1. Next, in the 19th clock from the 20th clock to the 38th clock of the clock pulse signal CK1, it rises in synchronization with the 20th clock down edge of the clock pulse signal CK1 and in synchronization with the 29th clock down edge of the clock pulse signal CK1. And stand up. Then, after the 39th clock, the same operation as the 20th to 38th clocks is repeated.

The pulse signal L output from the output terminal P8
D1 is output to each processor element PE of the processor unit 3100, and each processor element PE
While transferring each distortion calculated in step 1 to the adjacent left processor element PE,
In the case of the processor element PE (0, y) in the first column, it is controlled so that it is transferred to the similar block specifying unit 4000.

The pulse signal LD1 has a pulse width twice the pulse width of the clock pulse signal CK1 and is output so as to rise in synchronization with the 18th clock down edge of the clock pulse signal CK1. Next, in the 19th clock from the 20th clock to the 38th clock of the clock pulse signal CK1, the clock pulse signal CK
It is output so that it rises in synchronization with the down edge of the 29th clock of 1 and rises in synchronization with the down edge of the 38th clock of the clock pulse signal CK1. Then, after the 39th clock, the same operation as the 20th to 38th clocks is repeated.

The pulse signal L output from the output terminal P9
D2 is output to the similar block specifying unit 4000 and sets an initial state of an operation for obtaining the minimum distortion from the distortion calculated by the processor unit 3100. The pulse signal LD2 has a pulse width twice the pulse width of the clock pulse signal CK1 and is output so as to rise in synchronization with the down edge of the pulse signal LD1.

The pulse signal SMV output from the output terminal P10 is output to the similar block specifying unit 4000, and the minimum distortion MinDis and the motion vector M specified by the similar block specifying unit 4000.
V is controlled so as to be output to the outside of this device. The pulse signal SMV has a pulse width that is twice the pulse width of the clock pulse signal CK1, and at the 19th clock from the 20th clock to the 38th clock of the clock pulse signal CK1, the down edge of the 23rd clock of the clock pulse signal CK1. Is output in synchronization with the rising edge of the clock pulse signal CK1 and the rising edge of the clock pulse signal CK1 in synchronization with the down edge of the 34th clock. After the 39th clock, the same operation as the 20th to 38th clocks is repeated.

Further, the window data transfer unit 3
The pixel data of the current image block and the pixel data of the search window, which are input to 000, will be described. As shown in FIGS. 8 to 12, the pixel data of the current image block is synchronized with the down edge of the clock pulse signal CK1 by the system control unit 6000 in the current image memory 1000.
From the output terminal R, and is input to each processor element PE at the same time in synchronization with the rising edge of the next pulse.

From the 10th clock to the 18th clock of the clock pulse signal CK1, the current image block 211
Pixel data of a (0,0), a (0,1), a (0,2), a (1,2), a (1,1), a (1,0), a (2 , 0), a (2,1), a (1,0), are simultaneously input to each processor element PE in the ascending order, the descending order, and the ascending order in the vertical direction of the current image for each column.

Next, the pixel data of the current image block 212 is pixel data a (3,0), a (3 at the 20th clock to the 38th clock of the clock pulse signal in synchronization with each pulse from the 21st clock. , 1), a (3, 2), ... For each column in the vertical direction of the current image, the ascending order, the descending order, and the ascending order are input, and the current image block 2 shown in FIG.
Pixel data of the current image blocks 203 adjacent to each other in the 12 horizontal scanning directions are continuously input to each processor element PE in the order of descending order, ascending order, and descending order for each column. After the 39th clock, the pixel data of the current image block adjacent to each other in the horizontal scanning direction is sequentially input to each processor element PE, similarly to the 20th clock to the 38th clock.

The pixel data of the search window is also
The system control unit 6000 reads from the reference image memory 2000 through the output terminals S0 and S1 in synchronization with the down edge of the pulse of the clock pulse signal CK1, and in synchronization with the rising edge of the next pulse, the input register IR (3 , 0) and the input register IR (3, 3).

Of the pixel data in the search window 401, the pixel data whose vertical position y is 2 or less (y = 0, 1, 2) is read out through the output terminal S0 of the reference image memory 2000, and the clock pulse signal CK1. Pixel clocks b (0,2), b (0,1), b (0,0), b (1,0) in synchronization with each pulse from the first clock to the 18th clock , b (1,1), b (1,2), b (2,2), b (2,1), b (2,0), ... Descending, ascending, descending,
Are input to the input register IR (3,0) in this order.

Of the pixel data in the search window 401, the pixel data whose vertical position y is 3 or more (y = 3, 4) is the pixel data read through the output terminal S0 with respect to b (x, y). Pixel data b (x, y + 3) are simultaneously read out through the output terminal S1, and the input register I
Input to R (3,3). Next, the pixel data of the search window 401 is pixel data b (3,0) in synchronization with each pulse from the 21st clock at the 20th clock to the 38th clock of the clock pulse signal.
And b (3,3) for each column, in the order of ascending order, descending order, ascending order, ... In the vertical direction of the reference image, the search window 401 shown in FIG. The pixel data of the search window 402 shifted in the direction is sequentially input.

From the 39th clock onward, similar to the 20th clock to the 38th clock, a part of the pixel data of the pixel data of the previously input search window is re-input, and M pixels are shifted in the horizontal scanning direction. The pixel data of the searched window is input register IR
(3, 0) and the input register IR (3, 3) are sequentially input.

Next, the window data transfer unit 30
00 processor elements PE and registers V
S and IR will be described. First, the terminal arrangement and block diagram of each processor element PE (x, y) of the processor unit 3100 will be described. As shown in FIG. 15, each processor element PE (x, y) has input terminals X, YUi, YDi, YLi, YRi, Di and output terminals YUo, YDo, YLo, YRo, Do,
Further, it has input terminals respectively connected to output terminals P1, P2, P3, P4, P6, P7 and P8 of the system control unit 6000 (not shown).

As shown in FIG. 16, each processor element PE (x, y) has a transfer direction switching unit 360.
0, distortion calculation unit 3700 and distortion transfer unit 3800. The transfer direction switching unit 3600 is further composed of a selector 3610 and a flip-flop 3620.

The selector 3610 has input terminals S0, S
1, S2, A, B, C, D and an output terminal Y, and based on the signals input through the input terminal S0, the input terminal S1 and the input terminal S2, the input terminal A, the input terminal B,
The switching device connects any one of the input terminal C and the input terminal D to the output terminal Y. When the signal input through the input terminal S2 is 1, the input terminal D and the output terminal Y are connected regardless of the states of the signals input through the input terminal S0 and the input terminal S1. Further, when the signal input through the input terminal S2 is 0, and when the signals input through the input terminals S0 and S1 respectively represent (0, 0), the input terminal A and the output terminal Y are (1, 0 ), The input terminal B and the output terminal Y are connected, and when (0, 1) and (1, 1), the input terminal C and the output terminal Y are connected.

The flip-flop 3620 is a D flip-flop, has input terminals S and A and an output terminal Y, and in synchronization with the pulse of the signal input through the input terminal S, the data input to the input terminal A. Is latched to the output terminal Y. Input terminal A of selector 3610
Is an output terminal Y of another processor element PE (x, y−1) located one upper side via an input terminal YDi.
It is electrically connected to the output terminal of Do or the vertical side register VS (x, -1). Input terminal B is input terminal Y
Via Ui, it is electrically connected to the output terminal YUo of another processor element PE (x, y + 1) located at the one lower side or the output terminal of the vertical side register VS (x, 3). The input terminal C is connected via the input terminal YLi to another processor element PE (x +
1, y) output terminal YLo or input register IR
It is electrically connected to the output terminal of (3, y). The input terminal D is electrically connected to the output terminal YRo of another processor element PE (x-1, y) located on the left side or the output terminal of the horizontal side register HS (-1, y) via the input terminal YRi. Connected.

Further, the input terminal S0 of the selector 3610
Is electrically connected to the output terminal P3 of the system control unit 6000, the input terminal S1 is electrically connected to the output terminal P4 of the system control unit 6000, and the input terminal S2 is the output terminal P of the system control unit 6000.
6 electrically connected. The input terminal A of the flip-flop 3620 has the same processor element PE (x,
y) is electrically connected to the output terminal Y of the selector 3610, and the input terminal S is electrically connected to the output terminal P1 of the system control unit 6000.

Distortion calculation unit 3700
Is a subtractor 3710, a positive number converter 3720, an adder 3730, a flip-flop 3740, and an inverter 37.
50 and an AND operator 3760. The subtractor 3710 has input terminals A and B and an output terminal Y, and subtracts the data input through the input terminal B from the data input through the input terminal A and outputs the result through the output terminal Y.

The positive number converter 3720 has an input terminal A and an output terminal Y, converts data input through the input terminal A into positive number data by absolute value calculation, and outputs the output terminal Y.
Is output through. The adder 3730 has input terminals A and B and an output terminal Y, and adds the data input through the input terminal A and the data input through the input terminal B and outputs the result through the output terminal Y.

The inverter 3750 has an input terminal S and an output terminal Y, and the signal input through the input terminal S is 0.
In the case of, the data in which all the bits are represented by 1 is output through the output terminal Y, and when the input signal is 1, the data in which all the bits are represented by 0 are output through the output terminal Y. . The logical product operator 3760 is
It has input terminals A and B and an output terminal Y, performs a logical AND operation for each bit by the respective data input through the input terminal A and the input terminal B, and outputs the resulting data through the output terminal Y. It is a thing.

The input terminal A of the subtractor 3710 is the flip-flop 3 of the same processor element PE (x, y).
620 is electrically connected to the output terminal Y, and the input terminal B is electrically connected to the output terminal R of the current image memory 1000 via the input terminal X. The input terminal A of the positive number converter 3720 is electrically connected to the output terminal Y of the subtractor 3710 of the same processor element PE (x, y).

The input terminal A of the adder 3730 is the positive number converter 3720 of the same processor element PE (x, y).
Of the logical product operator 3 of the same processor element PE (x, y).
It is electrically connected to the output terminal Y of 760. The input terminal A of the flip-flop 3740 is electrically connected to the output terminal Y of the adder 3730 of the same processor element PE (x, y), and the input terminal S is electrically connected to the output terminal P1 of the system control unit 6000. Connected.

The input terminal S of the inverter 3750 is electrically connected to the output terminal P7 of the system control unit 6000. The input terminal A of the logical product operator 3760 is electrically connected to the output terminal Y of the inverter 3750 of the same processor element PE (x, y), and the input terminal B of the same processor element PE (x, y). Flip flop 3
It is electrically connected to the output terminal Y of 740.

The following operations are performed by positive number converter 3720, inverter 3750, AND operator 3760, adder 3730 and flip-flop 3740. here,
Clock pulse signal CK from flip-flop 3740
The data latched in synchronization with 1 is called the previous data, and the data output from the positive number converter 3720 is called the current data.

When the pulse signal CL input through the input terminal S of the inverter 3750 is 1, the adder 3730
Then, since the data input through the input terminal B becomes 0, the current data input through the input terminal A is output as it is through the output terminal Y and is input to the input terminal A of the flip-flop 3740. On the other hand, the inverter 37
When the pulse signal CL input to the input terminal S of 50 is 0, the previous data output from the output terminal Y of the flip-flop 3740 is input to the logical AND calculator 3760 through the input terminal B and through the output terminal Y. It is output and input to the input terminal B of the adder 3730. Therefore, in the adder 3730, the current data input through the input terminal A and the previous data latched by the flip-flop 3740 and input through the input terminal B are added and output through the output terminal Y. It is input to the input terminal A of the flip-flop 3740.

That is, when the pulse signal CL is 1, the integrated data of the adder 3730 is reset, and when the pulse signal CL is 0, the adder 3730 performs the integrating process of the previous data and the current data. The distortion transfer unit 3800 is further composed of a selector 3810 and a flip-flop 3820.

The selector 3810 has input terminals S, A,
B is a switching device that has an output terminal Y, and connects one of the input terminal A and the input terminal B to the output terminal Y based on a signal input through the input terminal S. Here, when the signal input through the input terminal S is 0, the input terminal A and the output terminal Y are connected, and when the signal input through the input terminal S is 1, the input terminal B and the output terminal Y are connected. Connected.

The flip-flop 3820 is composed of a D flip-flop, has input terminals S and A and an output terminal Y, and in synchronization with the pulse of the signal input through the input terminal S, the data input to the input terminal A. Is latched to the output terminal Y. Input terminal A of selector 3810
Is electrically connected via an input terminal Di to an output terminal Y of a flip-flop 3820 of another processor element PE (x + 1, y) located on the right side, and an input terminal B is connected to the same processor element PE ( (x, y) is electrically connected to the output terminal Y of the flip-flop 3740, and the input terminal S is electrically connected to the output terminal P8 of the system control unit 6000.

Input terminal A of flip-flop 3820
Is electrically connected to the output terminal Y of the selector 3810 of the same processor element PE (x, y), and the input terminal S is electrically connected to the output terminal P2 of the system control unit 6000. Next, the terminal arrangement and block diagram of each vertical side register VS (x, y) of the first vertical side register unit 3200 and the second vertical side register unit 3300 will be described.

As shown in FIG. 17A, each vertical side register VS (x, y) has an input terminal YUi, YDi,
YLi, YRi and output terminals YUo, YDo, YL
O and YRo, and further has input terminals respectively connected to output terminals P1, P3, P4 and P5 of a system control unit 6000 (not shown). FIG. 17 (b)
, Each vertical side register VS (x, y)
Is a selector 3611 and a flip-flop 3621
Consists of

The selector 3611 has input terminals S0, S
1, S2, A, B, C, D and an output terminal Y, and based on the signals input through the input terminal S0, the input terminal S1 and the input terminal S2, the input terminal A, the input terminal B,
The switching device connects any one of the input terminal C and the input terminal D to the output terminal Y. When the signal input through the input terminal S2 is 1, the input terminal D and the output terminal Y are connected regardless of the states of the signals input through the input terminal S0 and the input terminal S1. When the signal input through the input terminal S2 is 0, and when the signals input through the input terminals S0 and S1 respectively represent (0, 0), the input terminal A and the output terminal Y are connected to (1, 0 ), The input terminal B and the output terminal Y are connected, and when (0, 1) and (1, 1), the input terminal C and the output terminal Y are connected.

The flip-flop 3621 is a D flip-flop, has input terminals S and A, and an output terminal Y, and in synchronization with the pulse of the signal input to the input terminal S, the data input to the input terminal A. Is latched to the output terminal Y. The input terminal A of the selector 3611 is
Via the input terminal YDi, the output terminal YDo of another vertical side register VS (x, y−1) located one above or the output terminal Y of the processor element PE (x, 2).
It is electrically connected to Do. Input terminal B is input terminal Y
Via Ui, it is electrically connected to the output terminal YUo of another vertical side register VS (x, y + 1) located one lower side or the output terminal YUo of the processor element PE (x, 0). The input terminal C is connected via the input terminal YLi to another vertical side register VS located on the right side.
Output terminal YLo of (x + 1, y) or input register I
It is electrically connected to the output terminal YLo of R (3, y).
The input terminal D is, via the input terminal YRi, the output terminal YRo of another vertical side register VS (x-1, y) located on the left side or the horizontal side register HS (-1,
It is electrically connected to the output terminal of y). The input terminal S0 is the output terminal P3 of the system control unit 6000.
, The input terminal S1 is electrically connected to the output terminal P4 of the system control unit 6000, and the input terminal S2 is electrically connected to the output terminal P6 of the system control unit 6000.

Input terminal A of flip-flop 3621
Is electrically connected to the output terminal Y of the selector 3611 of the same vertical side register VS (x, y), and the input terminal S
Are electrically connected to the output terminal P1 of the system control unit 6000. Next, the input register unit 340
The terminal arrangement and block diagram of each input register IR (x, y) of 0 will be described.

As shown in FIG. 18A, each input register IR (x, y) has input terminals YUi, YDi and output terminals YUo, YDo, YLo, and further outputs of a system control unit 6000 (not shown). Terminals P1, P3
Input terminals respectively connected to. FIG.
As shown in (b), each input register IR (x, y)
Is a selector 3612 and a flip-flop 3622.
Consists of

The selector 3612 has input terminals S, A, B
And a output terminal Y, and based on a signal input through the input terminal S, connects one of the input terminal A and the input terminal B to the output terminal Y. Here, the signal input through the input terminal S is 0
When, the input terminal A and the output terminal Y are connected, and when the signal input through the input terminal S is 1, the input terminal B and the output terminal Y are connected.

The flip-flop 3622 is a D flip-flop, has input terminals S and A and an output terminal Y, and in synchronization with the pulse of the signal input through the input terminal S, the data input to the input terminal A. Is latched to the output terminal Y. Input register IR (3,-
2), input register IR (3, -1), input register I
The input terminal B of the selector 3612 of each of the R (3,1) and the input register IR (3,2) is an input terminal to the output terminal YUo of another input register IR (x, y + 1) located one below. It is electrically connected via YUi.

The input terminal A and the input terminal B of the selector 3612 of the input register IR (3,0) are electrically connected to the output terminal S0 of the reference image memory 2000 via the input terminals YDi and YUi, respectively. . Input terminal A of selector 3612 of input register IR (3,3)
And input terminal B are input terminals YDi and Y, respectively.
It is electrically connected to the output terminal S1 of the reference image memory 2000 via Ui.

The input terminal A of the selector 3612 of each of the input register IR (3,1), the input register IR (3,2) and the input register IR (3,4) is moved upward by one via the input terminal YDi. Another input register I located
It is electrically connected to the output terminal YDo of R (x, y-1). Selector 3612 of each input register IR (3, y)
The input terminal S of is electrically connected to the output terminal P3 of the system control unit 6000.

The input terminal A of the flip-flop 3622 of each input register IR (3, y) is the same as the input register I.
The input terminal S is electrically connected to the output terminal Y of the R (3, y) selector 3612, and the input terminal S is connected to the system control unit 60.
00 is electrically connected to the output terminal P1. Next, the terminal arrangement and block diagram of each horizontal side register HS (x, y) of the horizontal side register unit 5000 will be described.

As shown in FIG. 19A, each horizontal side register HS (x, y) has input terminals YLi, YRi and output terminals YLo, YRo, and an output terminal of a system control unit 6000 (not shown). P1, P5
It has input terminals respectively connected to P6. FIG.
As shown in FIG. 9 (b), each horizontal side register HS
(X, y) is composed of a selector 3613 and a flip-flop 3623.

The selector 3613 has input terminals S0, S
1, A, B, C and an output terminal Y, and an input terminal S0
And the signal S input through the input terminal S1
It is a switcher that connects any one of the input terminal A, the input terminal B, and the input terminal C to the output terminal Y based on H and SR. Here, when the signals SH and SR respectively inputted through the input terminals S0 and S1 are (0,0), the input terminal A and the output terminal Y are electrically connected, and when the signals SH and SR are (1,0), the input is made. The terminal B and the output terminal Y are electrically connected, and when (0, 1) and (1, 1), the input terminal C and the output terminal Y are electrically connected.

The flip-flop 3623 is composed of a D flip-flop, has input terminals S and A and an output terminal Y, and in synchronization with a pulse of a signal inputted through the input terminal S, data inputted to the input terminal A. Is latched to the output terminal Y. Input terminal A of selector 3613
Are electrically connected to the output terminal Y of the flip-flop 3623 of the same horizontal side register HS (x, y).
The input terminal B is, via the input terminal YLi, the output terminal YLo of another horizontal side register HS (x + 1, y) located on the right side, the processor element PE (0, y) or the vertical side register VS (0, y) output terminal YL
electrically connected to o. The input terminal C is the input terminal YR
It is electrically connected to the output terminal YRo of another horizontal side register HS (x-1, y) located on the left side through i. The input terminal S0 is the system control unit 6
000 output terminal P5 electrically connected to the input terminal S
1 is electrically connected to the output terminal P6 of the system control unit 6000.

Input terminal A of flip-flop 3623
Is electrically connected to the output terminal Y of the selector 3613 of the same horizontal side register HS (x, y), and the input terminal S
Are electrically connected to the output terminal P1 of the system control unit 6000. Next, the similar block identification unit 4
000 will be described in detail. As shown in FIG.
The similar block identifying unit 4000 includes a minimum distortion detecting unit 4100, a motion vector vertical component detecting unit 4200, and a motion vector horizontal component detecting unit 4300.

Minimum distortion detection unit 410
0 inputs a plurality of distortions calculated by the processor unit 3100 and detects the distortion having the minimum value from the input distortions. The motion vector vertical component detection unit 4200 specifies and identifies a candidate block based on the position information (row position) of the processor element PE (x, y) for which the minimum distortion detected by the minimum distortion detection unit 4100 is calculated. The motion vector vertical component MVy corresponding to the candidate block is identified.

Motion vector horizontal component detection unit 430
0 specifies a candidate block based on the position information (column position) of the processor element PE (x, y) for which the minimum distortion detected by the minimum distortion detection unit 4100 is calculated, and corresponds to the specified candidate block. The motion vector vertical component MVx is specified. The above units will be described below.

Minimum distortion detection unit 410
0 is further the comparator 4101 and the logical sum operator 410
2, comparator 4103, selector 4104, flip-flop 4105, flip-flop with selector 4106
Composed of The comparator 4101 has an input terminal A
It has 0, A1, A2 and output terminals M, Y. The input terminals A0, A1 and A2 are respectively the processor element PE (0, 0) and the processor element PE (0, 1).
And an output terminal Do of the processor element PE (0, 2). The output terminal Y is a comparator 41
03 input terminal A and selector 4104 input terminal B
Electrically connected to. The output terminal M is electrically connected to the input terminal B of the selector 4211 of the motion vector vertical component detection unit 4200.

The comparator 4101 inputs the distortions simultaneously output from the processor element PE (0, y) through the input terminals A0, A1 and A2, respectively, and selects the minimum value from the three distortions input. The distortion LMDis having the same is output through the output terminal Y. Also, the data LMVy indicating the input terminal to which the minimum distortion is input is output from the output terminal M. Here, the data L output from the output terminal M
MVy is 0 when the input terminal to which the minimum distortion is input is the input terminal A0, 1 when the input terminal is A1, and A
When it is 2, it represents 2.

The logical sum operator 4102 has input terminals A and B.
And an output terminal Y. The input terminal A is electrically connected to the output terminal P9 of the system control unit 6000, and the input terminal B is electrically connected to the output terminal Y of the flip-flop 4105. The output terminal Y is a comparator 41
03 is electrically connected to the input terminal B. The logical sum operator 4102 inputs the pulse signal LD2 output from the system control unit 6000 through the input terminal A, inputs the data output from the flip-flop 4105 through the input terminal B, and outputs the input signal LD2.
Is expressed by a bit string and the data input through the input terminal B is ORed, and the operation result is set to the output terminal Y. Here, when the signal LD2 is 0, all the bits of the data corresponding to the signal LD2 are represented by 0, and the data input from the input terminal B is output as the logical sum operation result. On the other hand, when the signal LD2 is 1, all the bits of the data corresponding to the signal LD2 are represented by 1, and this data is output as it is through the output terminal Y as the maximum value.

Comparator 4103 has input terminals A and B and output terminal Y. The input terminal A is electrically connected to the output terminal Y of the comparator 4101, and the input terminal B is electrically connected to the output terminal Y of the logical sum operator 4102. The output terminal Y is the input terminal S of the selector 4104 and the selector 4211 of the motion vector vertical component detection unit 4200.
Of the selector 4302 of the motion vector horizontal component detection unit 4300.

The comparator 4103 inputs the distortion output from the comparator 6101 through the input terminal A, inputs the data output from the logical sum operator 4102 through the input terminal B, and inputs the data through the input terminal A. When the distortion is larger than the data input through the input terminal B, the output terminal Y outputs the data Min representing 0, and the distortion input through the input terminal A is the data input through the input terminal B. If it is smaller, the data Min representing 1 is output from the output terminal Y.

The selector 4104 has input terminals S, A and B.
And an output terminal Y. The input terminal S is a comparator 41
03 is electrically connected to the output terminal Y, and the input terminal A is
The output terminal Y of the flip-flop 4105 is electrically connected, and the input terminal B is electrically connected to the output terminal Y of the comparator 4101. The output terminal Y is a flip-flop 4
It is electrically connected to the input terminal A of 105.

The selector 4104 inputs the data Min output from the comparator 4103 through the input terminal S,
A switching device that electrically connects one of the input terminals A and B to the output terminal Y based on the input data Min. When the data Min is 0, the input terminal A
And the output terminal Y are electrically connected, and when the data Min is 1, the input terminal B and the output terminal Y are electrically connected.

The flip-flop 4105 is a D flip-flop and has input terminals S and A and an output terminal Y. The input terminal S is the system control unit 6000.
Of the selector 4104, the input terminal A is electrically connected to the output terminal Y of the selector 4104, and the output terminal Y is the input terminal A of the selector 4104 and the input terminal B of the logical sum operator 4102. And an input terminal I of the flip-flop with selector 4106.

The flip-flop 4105 inputs the pulse signal CK2 output from the system control unit 6000 through the input terminal S, and inputs the pulse signal CK
The data input through the input terminal A is latched at the output terminal Y in synchronization with the rising edge of each pulse 2. The flip-flop with selector 4106 is configured by the flip-flop with selector 4400 shown in FIG. Here, as shown in FIG. 21, the flip-flop with selector 4400 has input terminals E, F, and I and an output terminal O, and is further composed of a selector 4401 and a flip-flop 4402. The input terminal E is electrically connected to the output terminal P2 of the system control unit 6000, and the input terminal F is the system control unit 6000.
Is electrically connected to the output terminal P10.

The selector 4401 has input terminals S, A, B
And an output terminal Y, the input terminal S is electrically connected to the input terminal E, and the input terminal A is the flip-flop 4
402 is electrically connected to the output terminal Y of the input terminal 402 and is connected to the input terminal B
Are electrically connected to the input terminal I. Selector 440
Reference numeral 1 denotes a switching device that electrically connects either one of the input terminal A and the input terminal B to the output terminal Y based on the pulse signal SMV input through the input terminal E and the input terminal S. When the signal input through the terminal S is 0, the input terminal A and the output terminal Y are electrically connected, and when the signal input through the input terminal S is 1, the input terminal B and the output terminal Y are electrically connected. Connect to each other.

Flip-flop 4402 is formed of a D flip-flop and has input terminals S and A and an output terminal Y. The input terminal S is electrically connected to the input terminal F, the input terminal A is electrically connected to the output terminal Y of the selector 4401, the output terminal Y is electrically connected to the output terminal O, and It is electrically connected to the input terminal A of the selector 4401.

The flip-flop 4402 inputs the pulse signal CK2 output from the system control unit 6000 through the input terminal F and the input terminal S, and inputs it to the input terminal A in synchronization with the rising edge of each pulse of the input pulse signal CK2. The input data is latched in the output terminal Y. The input terminal I of the flip-flop with selector 4106 is electrically connected to the output terminal Y of the flip-flop 4105. Flip-flop with selector 4106
Inputs the signal SMV output from the system control unit 6000 through the input terminal E, and the input signal S
The connection between the input terminal and the output terminal of the selector 4401 is switched based on the MV, and the data selectively input from the input terminal I is output through the output terminal O.

That is, the minimum distortion detection unit 4100 calls the data latched in the flip-flop 4105 in synchronization with the pulse signal CK2 the minimum distortion up to the previous time, and the data LMDis output from the comparator 4101 is the minimum distortion of this time. When the signal input through the input terminal A of the logical sum operator 4102 is 0, the minimum distortion of this time input through the input terminal A by the comparator 4103 and the previous minimum distortion input through the input terminal B. The smaller distortion is compared with the minimum distortion, and the smaller distortion is flipped through the selector 4104.
5 is to be latched.

Further, the input terminal S of the logical sum operator 4102
When the signal input to 1 is 1, the data in which all the bits representing the maximum value are 1 are input to the comparator 4103 as the initial state of the minimum distortion detection processing, and are compared with the current minimum distortion. Finally, the minimum distortion latched in the flip-flop 4105 is selectively input by the flip-flop with a selector 4106 and output as the minimum distortion MinDis.

Motion vector vertical component detection unit 420
0 is a selector 4201, a flip-flop 4202,
It is composed of a conversion table 4203 and a flip-flop 4204 with a selector. Selector 4201
Has input terminals S, A, B and an output terminal Y. The input terminal S is the minimum distortion detection unit 410
0 is electrically connected to the output terminal Y of the comparator 4103,
The input terminal A is the output terminal o of the flip-flop 4202.
The input terminal B is electrically connected to the output terminal M of the comparator 4101 of the minimum distortion detection unit 4100. The output terminal Y is electrically connected to the input terminal A of the flip-flop 4202.

The selector 4201 inputs the data Min output from the comparator 4103 through the input terminal S,
It is a switcher that electrically connects one of the input terminals A and B to the output terminal Y based on the input data Min. When the data Min is 0, the input terminal A and the output terminal Y are connected. Are electrically connected, and when the data Min is 1, the input terminal B and the output terminal Y are electrically connected.

Flip-flop 4202 is formed of a D flip-flop and has input terminals S and A and an output terminal Y. The input terminal S is the system control unit 6000.
Is electrically connected to the output terminal P2 of the selector 4201, and the input terminal A is electrically connected to the output terminal Y of the selector 4201. The output terminal Y is electrically connected to the input terminal A of the conversion table 4203 and the input terminal A of the selector 4201. Connected.

The flip-flop 4202 inputs the pulse signal CK2 output from the system control unit 6000 through the input terminal S and inputs the input pulse signal CK.
The data input through the input terminal A is latched at the output terminal Y in synchronization with the rising edge of each pulse 2. The conversion table 4203 has an input terminal A and an output terminal Y,
The input terminal A is the output terminal Y of the flip-flop 4202.
The output terminal Y is electrically connected to the input terminal I of the flip-flop with selector 4204. The conversion table 4203 calculates the motion vector vertical component MV based on the data My input through the input terminal A.
The data representing y is looked up from the conversion table and output to the output terminal Y.

Flip-flop with selector 4204
Is composed of a flip-flop 4400 with a selector. The input terminal I is electrically connected to the output terminal Y of the conversion table 4203. The flip-flop with selector 4204 inputs the signal SMV output from the system control unit 6000 through the input terminal E, switches the connection between the input terminal and the output terminal of the selector 4401 based on the input signal SMV, and from the input terminal I Selectively input the motion vector vertical component MVy to the output terminal O
Output from

Motion vector horizontal component detection unit 430
0 is a counter 4301, a selector 4302,
It is composed of a flip-flop 4303, a conversion table 4304, and a flip-flop 4305 with a selector. The counter 4301 has input terminals CL and CK and an output terminal Qn. The input terminal CL is electrically connected to the output terminal P9 of the system control unit 6000, the input terminal CK is electrically connected to the output terminal P2 of the system control unit 6000, and the output terminal Qn is the input terminal of the selector 4302. Electrically connected to B.

When the pulse signal LD2 input through the input terminal CL represents 1, the counter 4301 resets the data CTx to 0 and outputs it through the output terminal Qn, and the signal input through the input terminal CL outputs 0. In this case, the data CTx is incremented by 1 in synchronization with the rising edge of the pulse of the pulse signal CK2 input through the input terminal CK and is output through the output terminal Qn.

The selector 4302 has input terminals S, A and B.
And an output terminal Y. The input terminal S is electrically connected to the output terminal Y of the comparator 4103 of the minimum distortion detection unit 4100, the input terminal A is electrically connected to the output terminal Y of the flip-flop 4303, and the input terminal B is a counter. It is electrically connected to the output terminal Qn of 4301. The output terminal Y is a flip-flop 430.
3 is electrically connected to the input terminal A.

The selector 4302 inputs the data Min output from the comparator 4103 through the input terminal S,
It is a switcher that electrically connects one of the input terminals A and B to the output terminal Y based on the input data Min. When the data Min is 0, the input terminal A and the output terminal Y are connected. Are electrically connected, and when the data Min is 1, the input terminal B and the output terminal Y are electrically connected.

Flip-flop 4303 is a D flip-flop and has input terminals S and A and an output terminal Y. The input terminal S is the system control unit 6000.
Is electrically connected to the output terminal P2 of the selector 4302, the input terminal A is electrically connected to the output terminal Y of the selector 4302, and the output terminal Y is electrically connected to the input terminal of the conversion table 4304 and the input terminal A of the selector 4302. Connected to.

The flip-flop 4303 inputs the pulse signal CK2 output from the system control unit 6000 through the input terminal S and inputs the pulse signal CK
The data input to the input terminal A is latched to the output terminal Y in synchronization with the rise of each pulse of 2. Conversion table 4
304 has an input terminal A and an output terminal Y, the input terminal A is electrically connected to the output terminal Y of the flip-flop 4303, and the output terminal Y is electrically connected to the input terminal I of the flip-flop with selector 4305. Connected to. The conversion table 4304 looks up the data representing the motion vector horizontal component MVx from the conversion table based on the data Mx input to the input terminal A, and outputs it to the output terminal Y.

Here, when the motion vector to be obtained is the left half distortion, the horizontal component motion vector MVx is looked up from the conversion table corresponding to the left half search window based on the input data Mx, and the obtained motion vector is obtained. In the case of the right half distortion, the horizontal component motion vector MVx is looked up from the conversion table corresponding to the right half search window based on the input data Mx.

Flip-flop with selector 4305
Is composed of a flip-flop 4400 with a selector. The input terminal I is electrically connected to the output terminal Y of the conversion table 4305. The flip-flop with selector 4305 inputs the signal SMV output from the system control unit 6000 through the input terminal E, switches the connection between the input terminal and the output terminal of the selector 4401 based on the input signal SMV, and from the input terminal I The selectively input motion vector horizontal component MVx is output from the output terminal O.

Next, the operation will be described. Hereinafter, this Example 1
The operation for obtaining a motion vector by the motion vector search device of FIG. 8 will be described based on the time charts shown in FIGS. First, the window data transfer unit 30
The operation of transferring the pixel data of the search window 401 by 00 and calculating the distortion by the processor unit 3100 will be described based on the time charts shown in FIGS. 8 and 9.

Pixel data of the search window in the window data transfer unit 3000 is stored in the transfer direction switching unit 3600 of each processor element PE and the selectors 3610 and 3611 of the respective registers.
3612, 3613 select the respective flip-flops 3620, 3621, 3622, 3623.
Is held.

As shown in FIGS. 8 and 9, the pulse signals SU, SL and SR change in synchronization with the down edge of each pulse of the clock pulse signal CK1, and therefore the connection of each selector 3610, 3611, 3612. Is selected, the pixel data in the search window is latched by the flip-flops 3620, 3621, 3623 in synchronization with the rising edge of the clock pulse signal CK1.

For example, in the input register IR (3, y) shown in FIG. 18, the pulse signal SU is 0 until the third clock of the clock pulse signal CK1, and the selector 36
Since 0 is input to 12 through the input terminal S, the input terminal A and the output terminal Y are connected, and the clock pulse signal C
The pixel data of the search window input through the input terminal A in synchronization with each pulse of K1 is flip-flop 3
Latch 622.

Next, the pulse signal SU becomes 1 before the fourth clock of the clock pulse signal CK1 is issued,
In the selector 3612, since 1 is input through the input terminal S, the input terminal B and the output terminal Y are connected, and the search window input through the input terminal B is synchronized with the fourth clock of the clock pulse signal CK1. The pixel data is latched in the flip-flop 3622.

Further, the pixel data latched by the flip-flop 3622 is output to the output terminals YDo, YUo, YLo.
Is output to another adjacent input register IR, processor element PE or vertical side register VS via. Similarly, in the transfer direction switching unit 3600 of each processor element PE shown in FIG.
The pixel data input to the input terminal selected in 610 is latched by the flip-flop 3620 in synchronization with each pulse of the clock pulse signal CK1, and the output terminals YDo, Y
It is output to another processor element PE, a vertical side register VS, or a horizontal side register HS which is adjacent to the processor element PE via Uo, YLo, and YRo.

In each vertical side register VS shown in FIG. 17, the pixel data input to the input terminal selected by the selector 3611 is latched by the flip-flop 3621 in synchronization with each pulse of the clock pulse signal CK1. , Output to the other processor element PE, the vertical side register VS, or the horizontal side register HS, which are adjacent to each other, via the output terminals YDo, YUo, YLo, and YRo.

Further, in each horizontal side register HS shown in FIG. 19, the pixel data input to the input terminal selected by the selector 3613 is the clock pulse signal CK1.
Is latched by the flip-flop 3623 in synchronism with each pulse of and is output to another adjacent processor element PE, vertical side register VS or horizontal side register HS via the output terminals YLo and YRo.

Therefore, the pixel data of the search window in the window data transfer unit 3000 shown in FIGS. 22 to 33, which will be described below, includes the flip-flop 3620 of each processor element PE, the flip-flop 3621 of each vertical side register VS, The states latched in the flip-flop 3622 of each input register IR and the flip-flop 3623 of each horizontal side register HS are shown.

First, as shown in FIG. 22, the pixel data b (0,2) read from the reference image memory 2000 by the system control unit 6000 is synchronized with the first clock of the clock pulse signal CK1. IR
Input to (3, 0). Next, as shown in FIG. 23, the pixel data b (0,2) is input to the input register IR (3, in synchronization with the second clock of the clock pulse signal CK1.
Input register IR (3, which is located one lower than 0)
It is transferred to 1). At the same time, the pixel data b (0,1) and the pixel data b (0,4) are input register I respectively.
It is input to R (3,0) and the input register IR (3,3).

Then, as shown in FIG. 24, the pixel data held in the respective input registers IR (3, y) in synchronization with the third clock of the clock pulse signal CK1 is located in the input register one below. Transferred to IR (3, y + 1). At the same time, pixel data b (0,0) and pixel data b (0,3) are input register IR (3,
0) and the input register IR (3,3).

Next, as shown in FIG. 25, the pixel data held in each input register IR is synchronized with the fourth clock of the clock pulse signal CK1 and the pixel data held in each input register IR is positioned one to the left of each processor element PE or each processor element PE. It is transferred to the vertical side register VS. At the same time, pixel data b
(1,0) and pixel data b (1,3) are input register IR (3,0) and input register IR, respectively.
Input to (3, 3).

Then, as shown in FIG. 26, at the 5th and 6th clocks of the clock pulse signal CK1, the pixel data held in each processor element PE and each register are sequentially located one upper side. The pixel data transferred to the processor element or the register and read from the reference image memory 2000 are simultaneously input to the input register IR (3,0) and the input register IR (3,3).

As described above, each processor element PE
And between each register, the pixel data of the search window is synchronized with the clock pulse signal CK1 to the left, bottom, bottom,
As shown in FIG. 27, the processor elements PE (0,0), PE (0 , 1), PE (0,2), PE (1,0), PE (1,1), PE (1,2), PE (2,0), PE (2,1), PE (2, 2) pixel data b (0,0), b (0,1), b (0,2), b (1,0), b (1,1), b (1,2), b ( 2,0), b (2,1), b (2,2) are input at the same time and the distortion calculation is started. The above operation is repeated until the 18th clock of the clock pulse signal CK1.

The clock pulse signal CK1 shown in FIG.
From the 10th clock to the 18th clock of, each processor element PE (x, y) performs the following arithmetic processing. First, each processor element PE is synchronized with the 10th clock of the clock pulse signal CK1.
At (x, y), each pixel data b (x, y) latched by the flip-flop 3620 is subtracted by the subtractor 3
The pixel data a (0,0) of the current image block 201 is input to the subtractor 3710 via the input terminal B while being input to the 710 via the input terminal A.

Next, the subtractor 3710 calculates b (x, y) -a (0,0), and the positive number converter 3720 calculates | b (x, y) -a (0,0) | local distortion. Is calculated and input to the adder 3730 through the input terminal A.

On the other hand, the pulse signal CL is input to the inverter 3720 through the input terminal S. This pulse signal C
Since L becomes 1 in synchronization with the 9th clock down edge of the clock pulse signal CK1, 1 is input to the input terminal S of the inverter 3720, and as a result, the logical product operator 3
Data 0 is output through the output terminal Y of the 760 and input to the adder 3730 through the input terminal B. That is, in the adder 3730, the local distortion input through the input terminal A and 0 input through the input terminal B are added, and | b (x, y) -a (0,0) | is transferred to the flip-flop 3740. Is output.

Next, at the 11th clock of the clock pulse signal CK1, the pixel data b (x, y + 1) is input to the subtractor 3710 via the input terminal A, and the pixel data a (0 , 1) is input. As a result, the subtractor 3710 calculates b (x, y + 1) -a (0,1), and the positive number converter 3720 converts it into | b (x, y + 1) -a (0,1) | It is input to the adder 3730 through the input terminal A.

On the other hand, in the inverter 3720, 0 is input through the input terminal S. Therefore, the flip-flop 3
The previous data | b (x, y) -a (0,0) | latched in 740 is output through the output terminal Y of the AND operator 3760, and input to the adder 3730 through the input terminal B.

Next, in the adder 3730, the input terminal A
The current data input through the input terminal B and the previous data input through the input terminal B are added, and | b (x, y) -a (0,0) | + | b (x, y + 1) -a (0 , 1) | is output to the flip-flop 3740.

Next, since the pulse signal CL is 0 after the twelfth clock of the clock pulse signal, as in the eleventh clock, in the logical product operator 3760, the previous data latched by the flip-flop 3740 is input terminal B. Is input through the input terminal A of the adder 3730.
This time's data input via the input terminal B and the previous data input via the input terminal B are added and output via the output terminal Y.

That is, 1 of the clock pulse signal CK1
Similar calculation processing is performed from the second clock to the 18th clock, and as a result, | b (x, y) -a (0,0) | + | b (x, y + 1) -a (0,1) | + | B (x, y + 2) -a (0,2) | + | b (x + 1, y + 2) -a (1,2) | + | b (x + 1, y + 1) -a (1,1) | | B (x + 1, y) -a (1,0) | + | b (x + 2, y) -a (2,0) | + | b (x + 2, y + 1) -a (2,1) | + | b (X + 2, y + 2) -a (2,2) | is calculated and output to the flip-flop 3740. Therefore, the distortion corresponding to the current image block 201 is calculated.

Here, the distortion corresponding to the current image block 211 calculated by the processor unit 3100 is called the first distortion, and DR1 (x,
y). Hereinafter, similarly, the current image block 2 which is adjacent to the current image block 201 in the horizontal scanning direction
The distortion corresponding to 02 is called the second distortion, and the left half distortion is DL2 (x, y).
, And the right half distortion is represented by DR2 (x, y).

Next, the operation of transferring the distortion calculated by the processor unit 3100 to the similar block specifying unit 4000 will be described with reference to the time chart of FIG. First, the clock pulse signal C
Since the pulse signal LD1 becomes 1 in synchronization with the down edge of the 18th clock of K1, the selector 3810 of each processor element PE (x, y) receives 1 as input through the input terminal S and outputs as input terminal B and output. The terminal Y is connected.

Next, in synchronization with the 19th clock of the clock pulse signal CK1, each processor element PE
In the (x, y) flip-flop 3740, DR1 (x, y) calculated by the adder 3730 is latched and output through the output terminal Y and input to the selector 3810 through the input terminal B. Next, since the pulse signal LD1 becomes 0 in synchronization with the 19th clock down edge of the clock pulse signal CK1, 0 is input through the input terminal S in the selector 3810 of each processor element PE (x, y). , Input terminal A and output terminal Y
Are connected. Also, until the 29th clock, LD
Since 1 is 0, this connection state is held.

Next, at the 20th clock of the pulse signal CK1, in synchronization with the pulse signal CK2, the flip-flop 3820 of each processor element PE (x, y) latches DR1 (x, y) and outputs it. It is output through Y, and is output through the output terminal Do of each processor element PE (x, y). At this time, the processor element PE (0,0) and the processor element P
E (0,1) and processor element PE (0,
3) output DR1 (0,0), DR1 respectively
(0,1) and DR1 (0,2) are transferred to the similar block identifying unit 4000, respectively.

In addition, other processor elements (x,
y), each adjacent processor element PE (x
DR1 (x + 1, y) output from (+1, y) is transferred. Next, at the 21st clock of the pulse signal CK2, the flip-flop 3820 of each processor element PE (x, y) synchronizes with the pulse signal CK2,
DR1 (x + 1, y) is latched, output through the output terminal Y, and output through the output terminal Do of each processor element PE (x, y).

At this time, the processor element PE
(0,0), processor element PE (0,1) and processor element PE (0,3) have DR1 (1,0), DR1 (1,1) and DR1 respectively.
(1, 2) is output, and the similar block identification unit 40
00, and DR1 (2,2
0), DR1 (2,1) and DR1 (2,2) are input.

Next, at the 22nd clock of the pulse signal CK1, in synchronization with the pulse signal CK2, the processor element PE (0, 0) and the processor element PE (0,
1) and the flip-flop 3820 of the processor element PE (0, 3), respectively, DR1 (2,
0), DR1 (2,1) and DR1 (2,2) are latched, output through the output terminal Do, and transferred to the similar block specifying unit 4000.

That is, all DR1 (x, y) calculated by the processor unit 3100 have been transferred to the similar block specifying unit 4000. In the first embodiment, the distortion calculated by each processor element PE is sequentially transferred between the processor elements PE located in the same row for each column, so that each processor element PE (0, y) determines a similar block specifying unit. However, the distortion may be transferred row by row to the processor elements in the same column and transferred to the similar block specifying unit 4000.

Next, the similar block specifying unit 4000
The operation for obtaining the minimum distortion MinDis and the motion vector MV will be described with reference to the time chart shown in FIG. Note that My in FIG. 13 represents the data latched in the flip-flop 4202 of the motion vector vertical component detection unit 4200, and Mx represents the motion vector horizontal component detection unit 4300.
Represents the data latched in the flip-flop 4303. Further, MVx, y represent the horizontal component and the vertical component of the motion vector MV, respectively.

First, the pulse signal LD2 is synchronized with the down edge of the 19th clock of the clock pulse signal CK1.
Therefore, the minimum distortion detection unit 4
In 100, regardless of the data input to the logical sum operator 4102 through the input terminal A and input through the input terminal B, that is, the data output through the output terminal Y of the flip-flop 4105 and input, The data whose bit is 1 is output through the output terminal Y as the maximum distortion value.

In the motion vector horizontal component detection unit 4300, when 1 is input to the counter 4301 through the input terminal CL, the internal data CTx of the counter 4301 is reset to 0. Next, in the minimum distortion detection unit 4100 at the 20th clock of the clock pulse signal CK1, DR1 (0,0), DR1 calculated by the processor unit 3100.
(0, 1) and DR1 (0, 2) are synchronized with the pulse signal CK2, and input terminals A0 and A to the comparator 4101, respectively.
1 and A2.

Next, the comparator 4101 compares the input data, selects the smallest distortion from the data, and outputs the minimum distortion LMDis through the output terminal Y. In addition, the comparator 4101
Then, the data LMVy corresponding to the terminal to which the minimum distortion is input is output through the output terminal M. Here, the minimum distortion is DR1 (0,
0), and LMVy = 0 is output as shown in FIG.

Next, since the data having all the bits of 1 is output from the logical sum operator 4102 as the maximum value, the minimum distortion LMDis = DR1 (0, 0,
0) and the maximum value input through the input terminal B are compared, and since DR1 (0, 0) is smaller, Min = 1 is output through the output terminal Y.

Next, in the selector 4104, since 1 is input through the input terminal S, the input terminal B and the output terminal Y are connected, and DR1 (0,0) is output to the flip-flop 4105 through the output terminal Y. . On the other hand, in the motion vector vertical component detection unit 4200 at the 20th clock of the clock pulse signal CK1, the selector 42
In 01, since Min = 1 is input to the input terminal S,
The input terminal B and the output terminal Y are connected, and LMVy = 0 is output to the flip-flop 4202 through the output terminal Y.

Further, the clock pulse signal CK1 is set to 20.
Clock vector motion vector horizontal component detection unit 430
0, the counter 4301 outputs the pulse signal CK2.
In synchronization with, the internal data CTx = 0 reset by the pulse signal LD2 input to the input terminal CL is output through the output terminal Qn. Then selector 4
In 302, since Min = 1 is input to the input terminal S, the input terminal B and the output terminal Y are connected and CTx = 0.
Is output to the flip-flop 4302 through the output terminal Y.

Next, the minimum distortion detection unit 4100 at the 21st clock of the clock pulse signal CK2.
In the flip-flop 4105, the D output from the selector 4104 is synchronized with the pulse signal CK2.
R1 (0, 0) is latched and output to the flip-flop with selector 4180 through the output terminal Y. In the comparator 4101, DR1 (1,0), DR1 (1,1), DR1 are synchronized with the pulse signal CK2.
(1, 2) are input through the input terminals A0, A1, A2, respectively, and as a result, the minimum distortion LMD
Input terminal A to which is = DR1 (1,2) is output through the output terminal Y and DR1 (1,2) is input
LMVy = 2 representing 2 is output through the output terminal M.

In the OR calculator 4102, since the pulse signal LD2 = 0 is input through the input terminal S, the data input through the input terminal B, that is,
The previous minimum distortion DR1 (0,0) output from the output terminal Y of the flip-flop 4105 is input through the output terminal B and output through the output terminal Y.

Next, in the comparator 4103, the input terminal A
This minimum distortion LMD input through
is = DR1 (1,2) is compared with the previous minimum distortion DR1 (0,0) input through the input terminal B, and it is determined that DR1 (0,0) is smaller, and the output terminal Y Through, Min = 0 is output. Next, in the selector 4104, since 0 is input to the input terminal S, the input terminal A and the output terminal Y are connected, and DR1
(0,0) is output through the output terminal Y to the flip-flop 41
It is output to 05.

That is, the minimum distortion in the processor elements PE in the first and second columns is obtained. On the other hand, the clock pulse signal CK
In the motion vector vertical component detection unit 4200 at the 21st clock of 1 in the flip-flop 4202,
The data 0 output from the selector 4201 is latched in synchronization with the pulse signal CK2, and M is output through the output terminal Y.
It is output to the conversion table 4203 with y = 0.

Next, in the conversion table 4203, the motion vector vertical component MVy is looked up from the conversion table based on the data My = 0 input through the input terminal A, and is output to the flip-flop with selector 4204 through the output terminal Y. . In addition, the selector 4201
Then, since Min = 0 output from the comparator 4103 is input, the input terminal A and the output terminal Y are connected, and the previous LMVy = latched in the flip-flop 4202 =
0 is input again through the input terminal A and is output to the flip-flop 4202 through the output terminal Y.

Further, the clock pulse signal CK1 21
Clock vector motion vector horizontal component detection unit 430
At 0, the flip-flop 4302 latches the data 0 output from the selector 4302 and outputs it to the conversion table 4304 as Mx = 0 through the output terminal Y. Next, in the conversion table 4304, based on the data Mx = 0 input through the input terminal A, the motion vector horizontal component M corresponding to the right half distortion is generated.
Vx is looked up from the conversion table and output terminal Y
To the flip-flop with selector 4305.

Further, in the counter 4301, the counted output count CTx is output as CTx = 1 through the count output terminal Qn in synchronization with the pulse signal of CK2. Next, in the selector 4302, since Min = 0 outputted from the comparator 4103 is inputted through the input terminal S, the input terminal A and the output terminal Y are connected, and the previous value CTx = 0 is again inputted through the input terminal A. It is input and output to the flip-flop 4302 through the output terminal Y.

Next, the minimum distortion detection unit 4100 of the 22nd clock of the clock pulse signal CK2.
In the flip-flop 4105, the D output from the selector 4104 is synchronized with the pulse signal CK2.
R1 (0, 0) is latched and output to the flip-flop with selector 4180 through the output terminal Y. Further, in the comparator 4101, DR1 (2,0), DR1 (2,1), DR1 are synchronized with the pulse signal CK2.
(2, 2) is input through the input terminals A0, A1, A2, respectively, and as a result, the minimum distortion LMD
Input terminal A to which is = DR1 (2,1) is output through the output terminal Y and DR1 (2,1) is input
LMVy = 1 representing 2 is output through the output terminal M.

In the OR calculator 4102, since the pulse signal LD2 = 0 is input through the input terminal S, the data input through the input terminal B, that is,
The previous minimum distortion DR1 (0,0) output from the output terminal Y of the flip-flop 4105 is input through the output terminal B and output through the output terminal Y.

Next, in the comparator 4103, the input terminal A
This minimum distortion LMD input through
is = DR1 (2,1) is compared with the previous minimum distortion DR1 (0,0) input through the input terminal B, and it is determined that DR1 (2,1) is smaller, and the output terminal Y Through, Min = 1 is output. Next, in the selector 4104, since Min = 1 is input to the input terminal S, the input terminal B and the output terminal Y are connected, and DR1 (2,1) is output to the flip-flop 4105 through the output terminal Y.

That is, the minimum distortion D (2,1) is obtained from all DR1 (x, y) calculated by the processor unit 3100. on the other hand,
In the motion vector vertical component detection unit 4200 of the 22nd clock of the clock pulse signal CK1, in the flip-flop 4202, in synchronization with the pulse signal CK2,
The data 0 output from the selector 4201 is latched, My = 0 is set through the output terminal Y, and conversion table 4
It is output to 203.

Next, in the conversion table 4203, the motion vector MVy is looked up from the conversion table based on the data My = 0 input through the input terminal A, and is output to the flip-flop with selector 4204 through the output terminal Y. In the selector 4201,
Since Min = 1 output from the comparator 4103 is input, the input terminal B and the output terminal Y are connected, and the comparator 4
The current LMVy = 1 output from 101 is input through the input terminal B and output to the flip-flop 4202 through the output terminal Y.

Furthermore, 22 of the clock pulse signal CK1
Clock vector motion vector horizontal component detection unit 430
At 0, the flip-flop 4302 latches the data 0 output from the selector 4302 and outputs it to the conversion table 4304 as Mx = 0 through the output terminal Y. Next, in the conversion table 4304, based on the data Mx = 0 input through the input terminal A, the motion vector horizontal component M corresponding to the right half distortion is generated.
Vx is looked up from the conversion table and output terminal Y
To the flip-flop with selector 4305.

Further, in the counter 4301, the counted-up data CTx is synchronized with the pulse signal of the pulse signal CK2, and CTx = CTx = through the count output terminal Qn.
It is output as 2. Then, in the selector 4302,
Min = 1 output from the comparator 4103 is the input terminal S
Since this is input through the input terminal B and the output terminal Y, CTx = 2 of this time is input through the input terminal B and output to the flip-flop 4302 through the output terminal Y.

Next, the minimum distortion detection unit 4100 at the 23rd clock of the clock pulse signal CK2.
In the flip-flop 4105, the D output from the selector 4104 is synchronized with the pulse signal CK2.
R1 (2,1) is latched and output to the flip-flop with selector 4180 through the output terminal Y. On the other hand, in the motion vector vertical component detection unit 4200 at the 23rd clock of the clock pulse signal CK1, the flip-flop 4202 latches the data 1 output from the selector 4201 in synchronization with the pulse signal CK2 and outputs My through the output terminal Y. = 1 is output to the conversion table 4203.

Next, in the conversion table 4203, the motion vector MVy is looked up from the conversion table based on the data My = 1 input through the input terminal A, and is output to the flip-flop with selector 4204 through the output terminal Y. In addition, the clock pulse signal C
In the motion vector horizontal component detection unit 4300 at the 23rd clock of K1, the flip-flop 4302 latches the data 2 output from the selector 4302 and outputs it to the conversion table 4304 through the output terminal Y as Mx = 2.

Next, in the conversion table 4304, the motion vector horizontal component MVx corresponding to the right half distortion is looked up from the conversion table based on the data Mx = 0 input through the input terminal A, and through the output terminal Y, the flip-flop with selector is selected. 4305 is output. Further, since the pulse signal SMV becomes 1 in synchronization with the down edge of the 23rd clock of the clock pulse signal CK1, the minimum distortion detection unit 41
00, motion vector vertical component detection unit 4200, and motion vector horizontal component detection unit, respectively, with flip-flops 4106, 4204, 4305 with selectors
Then, 1 is input to the selector 4401 through the input terminal S via the input terminal E, respectively, and the input terminal B and the output terminal Y of the selector 4401 are electrically connected.

Next, the minimum distortion detection unit 4100 at the 24th clock of the clock pulse signal CK2.
, The minimum distortion DR output from the flip-flop 4105 is synchronized with the pulse signal CK2.
1 (2,1) is a flip-flop with selector 4106
Latched in the flip-flop 4402 of the output terminal O
Is output through.

On the other hand, the motion vector vertical component detection unit 4200 at the 24th clock of the clock pulse signal CK1
In the same manner, similarly, the motion vector vertical component MVy = 0 output from the conversion table 4203 is output through the output terminal O. Furthermore, the motion vector horizontal component detection unit 4 at the 24th clock of the clock pulse signal CK1
In 300, the motion vector horizontal component MVx = 2 similarly output from the conversion table 4305 is output through the output terminal O.

The pulse signal SMV is synchronized with the 24th clock down edge of the clock pulse signal CK1.
Is 0, so the minimum distortion detection unit 4
100, motion vector vertical component detection unit 4200, and motion vector horizontal component detection unit, respectively, with flip-flops 4106, 4204, 430 with selectors
5, the selector 4401 is input via the input terminal E, respectively.
0 is input to the selector 4401 through the input terminal S.
The input terminal A and the output terminal Y are electrically connected.

Therefore, the flip-flops 4106, 4204, 43 with selectors are selected from the 34th clock onward.
In 05, the minimum distortion DR1 (2,1), the motion vector vertical component MVy = 0, and the motion vector horizontal component MVx = 2 are repeatedly output. That is, all candidate blocks 3 in the search window 401
01 and the current image block 211, the minimum distortion MinDis (2,1) and the motion vector MV.
x, y = (2,0) is obtained.

Next, the first distortion DR1 (x,
y) At the time of calculation, the window data transfer unit 3000
The pixel data of the search window 401 output from the processor element PE (0, y) and the vertical side register SR (0, y) in the first column of the horizontal side register S of the horizontal side register unit 5000 are respectively received.
The operation of holding in R will be described based on the time chart shown in FIG.

First, the pulse signal SH is set to 1 in synchronization with the 10th clock down edge of the clock pulse signal CK1.
And the selector 3614 of each horizontal side register HS
Then, 1 is input through the input terminal S0, and the input terminal B
And the output terminal Y are electrically connected. Next, as shown in FIG. 28, each processor element PE (0, y) is synchronized with the 11th clock of the clock pulse signal CK1.
And pixel data b (0,0) and b (0,0) held in the vertical side registers VS (0, y), respectively.
1), b (0,2), b (0,3) and b (0,4)
Are the horizontal side registers HS (-
1, y) is latched in the flip-flop 3624.

Next, from the 12th clock to the 15th clock of the clock pulse signal CK1, both the pulse signals SH and SR are 0, and in the selector 3613 of each horizontal side register, 0 is input to the input terminals S0 and S1.
Is input, and the input terminal A and the output terminal Y are electrically connected. Therefore, the pixel data output through the output terminal Y of the flip-flop 3623 of each horizontal side register is input again through the input terminal A of the flip-flop 3624 of the same horizontal side register HS.

Next, as shown in FIG. 29, in synchronization with the 16th clock of the clock pulse signal CK1, the pixel data held in each horizontal side register HS (-1, y) is one to the left. Horizontal side register HS (-
2, y) and at the same time processor element P
The pixel data held in E (0, y) and the vertical side register (0, y) are respectively stored in the horizontal side register H.
It is transferred to S (-1, y).

Next, as shown in FIG. 30, in synchronization with the 17th clock of the clock pulse signal CK1, the pixel data held in each horizontal side register HS (-2, y) is one to the left. Horizontal side register HS (-
3, y), and at the same time, each horizontal side register H
The pixel data held in S (-1, y) are transferred to the left side horizontal register HS (-2, y), respectively, and at the same time, the processor element PE (0, y) and the vertical side register (0) are simultaneously transferred. , Y) are respectively transferred to the horizontal side registers HS (-1, y).

As a result, the pixel data in the search window 401 for M = 3 columns is stored in the horizontal side register unit 5.
It has been held at 000. Next, the search window 4 held in the horizontal side register unit 5000
The operation of returning the pixel data of 01 to the window data transfer unit again will be described based on the time chart of FIG.

The pulse signal SR becomes 1 in synchronization with the 18th clock down edge of the clock pulse signal CK1. Then, from the 19th clock to the 21st clock of the clock pulse signal CK1, the pixel data held in each horizontal side register HS is located one right side in synchronization with each pulse of the clock pulse signal CK1. The data is sequentially transferred to the horizontal side register HS, the processor element PE or the vertical side register VS.

More specifically, first, since the pulse signal SR becomes 1 in synchronization with the 18th clock down edge of the clock pulse signal CK1, each processor element PE (x, y) has an input terminal to the selector 3610. 1 is input through S2, and the input terminal D and output terminal Y of the selector 3610 are electrically connected. Also, in each vertical side register VS (x, y),
1 is input to the selector 3611 through the input terminal S2, and the input terminal D and output terminal Y of the selector 3611 are input.
And are electrically connected. Each horizontal side register VS
In (x, y), 1 is input to the selector 3613 through the input terminal S1, and the input terminal C and the output terminal Y of the selector 3613 are electrically connected.

Next, as shown in FIG. 31, the search window 401 held in each processor element PE, each vertical side register VS and each horizontal side register HS is synchronized with the 19th clock of the clock pulse signal CK1. The pixel data are transferred to the right side horizontal register HS, the processor element PE, or each vertical side register VS, respectively.

Next, as shown in FIGS. 32 and 34, in synchronization with the 20th clock and the 21st clock of the clock pulse signal CK1, the pixel data of the search window 401 are respectively one processor element PE on the right side and the vertical. It is sequentially transferred to the side register VS and the horizontal side register HS, and each processor element PE
And each vertical side register holds the same pixel data of the search window 401 as when the first distortion calculation was started.

Next, the operation of obtaining the motion vector MV by the pipeline processing in the motion vector search device of the first embodiment will be described based on the time charts shown in FIGS. As shown in FIG. 10, from the 21st clock to the 29th clock of the clock pulse signal CK1, the pixel data of the search window 401 returned from the horizontal side register unit 5000 corresponds to the current image block 212 shown in FIG. The second left half distortion is calculated by the processor unit 3100 in the same manner as the first right half distortion.

Here, the pixel data of the fourth and subsequent columns of the search window 401 are read out again from the reference image memory 2000 and supplied to the input register unit 3400. Then, the second left half distortion DL2 (x, y) calculated by the processor unit 3100.
Is transferred to the similar block identifying unit 4000.

Next, as shown in FIG. 14, from the 31st clock of the clock pulse signal CK1, the minimum distortion MinD is set by the similar block specifying unit 4000.
is = DL2 (1,0) and motion vector MVx, y
= (-2, -1) is obtained and output at the 35th clock. Since the motion vector MV obtained here is a motion vector corresponding to the left half search window corresponding to the current image block 212, the conversion table 43
It is converted into MVx corresponding to the search range of the left half by 04.

Further, as shown in FIG. 11, from the 30th clock to the 38th clock of the clock pulse signal CK1, M is compared with the search window 401 shown in FIG.
Search window 402 shifted horizontally by the number of pixels
A second right half distortion DR2 (x, y) of the current image block 212 and the current image block 212 is calculated, and part of the pixel data of the search window 402 is held in the horizontal side register unit 5000.

At this time, since the pixel data in the search window 402 is sequentially transferred in the downward, left, upward, left, and upward directions among the respective processor elements PE, the horizontal side register unit 5000 receives (M-
1) = 2 columns of pixel data may be held. Then
Pixel data for two columns held in each horizontal side register HS and the processor element PE in the first column
The pixel data held in (0, y) and the vertical side register VS (0, y) are respectively transferred to the right twice, and based on the returned pixel data of the search window 402 and the pixel data of the current image block 213. Then, the calculation of the third right half distortion is started.

That is, in the 19th clock from the 21st clock to the 38th clock, first, the right half search window 401 used for the first right half distortion calculation becomes the left half search window 401 corresponding to the current image block 212. The second left half distortion is calculated, and then the current image block 212
The second right half distortion is calculated from the right half search window 402 corresponding to, and pixel data of (M−1) columns of the pixel data of the right half search window 402 is the horizontal side register unit 5000.
, And then each horizontal register HS, as well as 1
The pixel data held in the processor element PE (0, y) in the column and the vertical side register VS (0, y) are returned to the distortion calculation start position.

Next, as shown in FIG. 34, after the 39th clock of the clock pulse signal CK1, the 19th clock is set as one cycle, and the distortions corresponding to the current image block adjacent in the horizontal scanning direction are left half and right respectively. The distortion is sequentially calculated from the half search window, and also in the similar block specifying unit 4000, the respective distortions calculated by the processor unit 3100 are transferred, and the minimum distortion MinDis and the motion vector MV are sequentially calculated, and the pipeline processing is realized. Has been done.

In the first embodiment, the pixel data held in the horizontal side register unit 5000, and
The pixel data held in the processor element PE (x, y) in the first column and the horizontal side register HS was transferred to the right as it is, and the next distortion was calculated. It can also be calculated by transferring in the reverse order of the distortion calculation. However, the same applies to the pixel data of the current image block in the opposite order.
Need to supply 100. This process can eliminate the loss of the return process, and is effective, for example, when the search range is tripled.

That is, as shown in FIG. 6, first, the distortion is calculated from the search window 401 and the current image block 211. Next, search window 40
2 and the current image block 211, the distortion corresponding to the current image 212 is calculated while returning the pixel data of the search window 402, and then the distortion between the search window 402 and the current image block 213 is calculated. .

Next, the distortion is calculated from the search window 403 and the current image block 212, then the distortion corresponding to the current image 213 is calculated while returning the pixel data of the search window 403, and then the search window 403 and the current image 213 are calculated. Image block 214
And the distortion is calculated, and thereafter, the same processing is sequentially performed.

Also, the similar block specifying unit 4000
It goes without saying that the distortion calculated by the processor unit 3100 is transferred, and the minimum distortion MinDis and the motion vector MV are sequentially obtained. As described above, according to the motion vector search device of the first embodiment, the pixel data of the search window and the pixel data of one current image block are used for a plurality of current image blocks sharing the same pixel data of the search window. Distortion is calculated, the pixel data of the search window discharged from the search window transfer unit 3000 at this time is held in the horizontal side register unit 5000, and the held pixel data of the search window is again searched for by the search window transfer unit 3.
Since the distortion corresponding to the different current image block can be calculated by returning to 000, the amount of data for accessing the pixel data of the search window to the reference image memory 2000 can be reduced, and the pixel data of the search window can be relatively obtained. Access time can be shortened. Therefore, the motion vector search process can be speeded up.

(Second Embodiment) FIG. 35 is a diagram showing a motion vector search device according to a second embodiment of the present invention. In the second embodiment, the characteristic part of the present invention will be specifically described. FIG.
As shown in FIG. 7, the motion vector search device of the second embodiment includes the current image cache memory 1001 between the current image memory 1000 and the window data transfer unit 3000 of the motion vector device of the first embodiment, and the reference image memory 2000. And window data transfer unit 3000
A reference image cache memory 2002 is provided between and.

The current image cache memory 1001 is composed of a volatile memory and has a system control unit 60.
00, the pixel data of the current image block read from the current image memory 1000 is stored, and the stored pixel data is output to the processor unit 3100. In addition, from the current image cache memory 1001 to the processor unit 310 of the window transfer unit 3000.
The transfer of pixel data to 0 is faster than the transfer of pixel data from the current image memory 1000 to the current image cache memory 1001.

The reference image cache memory 2001 is composed of a volatile memory, and has the system control unit 6
000 stores pixel data of the search window read from the reference image memory 2000, and
The stored pixel data is output to the input register unit 3400. Further, the transfer of pixel data from the reference image cache memory 1001 to the input register unit 3400 of the window transfer unit 3000 is performed at a higher speed than the transfer speed of pixel data from the reference image memory 2000 to the current image cache memory 2001.

Therefore, pixel data are read from the current image memory 1000 and the reference image memory 2000 in advance and stored in the current image cache memory 1001 and the reference image cache memory 2001, respectively.
When calculating the distortion, the pixel data of the current image block can be supplied to the processor unit 3100 at a higher speed by reading the pixel data from the cache memories 1001 and 2001 and supplying the pixel data to the processor unit 3100. Therefore, a faster distortion calculation process can be performed.

As described in the first embodiment, the pixel data of the current image block and some search windows are
Each is supplied to the window data transfer unit 3000 twice. On the other hand, in the second embodiment, each pixel data may be read from the current image memory 1000 and the reference image memory 2000 only once. Therefore, the current image memory 1000 and the reference image memory 2
Since 000 does not need to be configured with a high-speed memory, it can be configured with an inexpensive memory. Therefore, the cost of the memory can be reduced.

Generally, the pixel data of the search window has a larger number of data than the pixel data of the current image block. Therefore, the use of the cache memory is particularly effective. (Third Embodiment) FIGS. 36 and 37 are diagrams showing a motion vector search device according to a third embodiment of the present invention. In the third embodiment, a characteristic part of the present invention will be specifically described.

As shown in FIG. 38, the motion vector search device of the third embodiment includes the first vertical side register unit 320 of the window data transfer unit 3000 of the first embodiment.
The 0 and second vertical side register units are configured by the third vertical side register unit 3500, and the input register unit 3400 is configured by the input register unit 3401.

The vertical side register unit 3500 is
It is composed of six vertical side registers VS. The input register unit 3401 is composed of five input registers IR. Here, assuming that the processor element PE of the window data transfer unit 3001 and each register VS, IR are denoted by coordinates (x, y) as in the first embodiment, as shown in FIG. The input terminal YDi of the processor element PE (x, 0) in the first row is at the bottom end, that is,
The vertical side register VS (x, 5) of the fifth row is connected to the output terminal YDo of the vertical side register VS of the fifth row.
The input terminal YUi of (x, 5) is connected to the output terminal YUo of the processor element PE (x, 0) in the first row.

The input terminal YDi of the input register (3, 0) in the first row is the input register IR (3, 5) in the fifth row.
Connected to the output terminal YDo of the input register I of the fifth row
The input terminal YUi of R (3,5) is connected to the output terminal YUo of the input register IR (3,0) of the first row. That is, each processor element and each register are connected in a ring shape for each column.

Each processor element PE in the first line
The pixel data of the search window output from (x, 0) in the upward direction in FIG. 36 is transferred to the vertical side register VS (x, 5) on the fifth row and the vertical side register VS (( Pixel data output downward in the figure from (x, 5) is transferred to the processor element PE (x, 0) in the first row.

In addition, 1 of the input register unit 3421
Similarly, the pixel data of the search window is transferred to the input registers IR of the 5th and 5th rows. Therefore, in the first embodiment, the first and second vertical side register units 3200 are provided on both sides of the processor unit 3100, respectively.
In contrast to the arrangement of the 3300, in the second embodiment, since the third side register unit 3500 may be arranged on one side of the processor unit 3100, the circuit scale can be reduced by half.

When actually constructing the circuit, for example, as shown in FIG. 37, the distance between the processor elements PE and the vertical side registers VS is evenly arranged for each column to shorten the length. Since the transfer bus can be formed, it is possible to configure a stable circuit with few errors in the transfer operation and to reduce the circuit scale.

The operation for obtaining the motion vector MV is as follows:
It goes without saying that the operation is performed in the same manner as the first embodiment based on the time charts shown in FIGS. (Fourth Embodiment) FIG. 38 is a diagram showing a motion vector search device according to a fourth embodiment of the present invention. In the fourth embodiment, a characteristic part of the present invention will be specifically described.

As shown in FIG. 38, the motion vector search device is the search window transfer unit 300 of the second embodiment.
1, the processor element PE and the vertical side register VS are increased by one column. In the fourth embodiment, the number M of columns of the current image block is an even number, and N
= 3, M = 4, and the current image block is composed of pixel data in 3 rows and 4 columns. Further, it is assumed that the search window has H = 5 and L = 7 and is composed of pixel data of 5 rows and 7 columns. In this case, the search range is (H−N + 1) × (L−M + 1) = 12, and in the window transfer unit 3002, the distortions are calculated by the processor elements PE of 3 rows and 4 columns corresponding to 12 candidate blocks. It

The horizontal side register unit 5000 is
It is composed of (M-1) × N horizontal side registers. That is, when the number of columns M of the current image block is an even number, the pixel data of the search window is displayed at the start of distortion calculation and at the end of distortion calculation.
Held in each processor element, each vertical side register or each horizontal side register at the same vertical position,
The pixel data in the search window in the Mth column is 1
Since it is held in the processor element PE in the column and the vertical side register VS, the search window discharged from each processor element PE (0, y) in the first column and each vertical side register VS (0, y) when the distortion is calculated. The pixel data of is for (M-1) columns.

The operation for obtaining the motion vector MV is as follows:
It goes without saying that the same operation as in Example 1 and Example 2 is performed. In this way, when the number of columns of the current image block is even, one column is smaller than that in the case of odd numbers (M-1).
Since the vertical side register unit 5000 can be configured by the vertical side registers HS for columns, the circuit scale can be reduced.

[0249]

According to the first aspect of the invention, for a plurality of current image blocks sharing the same search window pixel data, the distortion is calculated from the search window pixel data and one current image block pixel data. Then, at this time, the pixel data of the search window discharged from the first and second pixel data transfer and holding means is held in the third pixel data transfer and holding means, and the held pixel data of the search window is again stored in the first and second pixel data transfer holding means. Since the distortion corresponding to different current image blocks can be calculated by returning to the second pixel data transfer means, the amount of data for accessing the pixel data of the desired search window from the pixel data of the reference image can be reduced. Can
The access time of the pixel data of the search window can be shortened relatively. Therefore, the motion vector search process can be speeded up.

According to the second aspect of the present invention, for a plurality of current image blocks sharing the same search window pixel data, the distortion is calculated from the search window pixel data and one current image block pixel data. Then, at this time, the pixel data of the search window discharged from the first and second pixel data transfer and holding means is held in the third pixel data transfer and holding means, and the held pixel data of the search window is again stored in the first and second pixel data transfer holding means. Since the distortion corresponding to the different current image block can be returned to the second pixel data transfer means, the pixel data of the desired search window is accessed to the reference image data storage means in which the pixel data of the reference image is stored. You can reduce the amount of data to It is possible to shorten the access time. Therefore, the motion vector search process can be speeded up.

According to the invention of claim 3, in the invention of claim 2, after the distortion calculation is completed by the distortion calculating means, the pixel data in the search window is returned to the state at the time of starting the distortion calculation. Since the distortion calculation corresponding to the current image block different from the current image block can be performed, the distortion can be continuously calculated for the two current image blocks. Therefore, the motion vector search process can be speeded up.

According to the fourth aspect of the present invention, for a plurality of current image blocks sharing the same search window pixel data, the distortion is calculated from the search window pixel data and one current image block pixel data. Then, at this time, the pixel data of the (M-1) th column search window discharged from the first and second pixel data transfer holding means is held in the third pixel data transfer holding means, and the held pixel data is held. Can be returned to the first and second pixel data transfer means again to calculate the distortions corresponding to different current image blocks. Therefore, a desired search window can be stored in the reference image data storage means in which the pixel data of the reference image is stored. It is possible to reduce the amount of data to access the pixel data of It is possible to shorten the access time. Therefore, the motion vector search process can be speeded up.

According to the invention described in claim 5, in the invention described in claim 4, when the pixel data in the search window returns to the position at the start of the distortion calculation after the distortion calculation, the transfer direction of the pixel data is changed again. Since the distortion calculation corresponding to different current image blocks can be started by switching to one of the row directions, the transfer direction can be easily switched, and the distortion calculation can be started quickly. .

According to the sixth aspect of the invention, for a plurality of current image blocks sharing the same search window pixel data, the distortion is calculated from the search window pixel data and one current image block pixel data. Then, at this time, the pixel data of the M columns of the search window discharged from the first and second pixel data transfer holding means is held in the third pixel data transfer holding means, and the held pixel data is again stored in the first pixel data. Since the distortion corresponding to the different current image block can be calculated by returning to the second pixel data transfer means, the pixel data of the desired search window can be stored in the reference image data storage means in which the pixel data of the reference image is stored. The amount of data to be accessed can be reduced, and the pixel data in the search window can be accessed relatively. It is possible to shorten the time. Therefore, the motion vector search process can be speeded up.

According to the invention described in claim 7, in the invention described in claim 6, when the pixel data in the search window returns to the position at the start of the distortion calculation after the distortion calculation, the transfer direction of the pixel data is changed again. Since the distortion calculation corresponding to different current image blocks can be started by switching to one of the row directions, the transfer direction can be easily switched, and the distortion calculation can be started quickly. .

According to the eighth aspect of the present invention, for a plurality of current image blocks sharing pixel data of the same search window, pixel data of a desired search window is read from the reference image data storage means and stored at high speed. The reference image data storage means is not accessed when the distortion corresponding to the current image block is calculated, and the high-speed transfer storage means is repeatedly accessed to store the search window in the first and second pixel data transfer holding means. Pixel data can be provided. Therefore, the pixel data in the search window can be transferred at high speed to the first and second pixel data transfer holding means, and
Since the reference image storage means does not need to be a high speed memory, it can be configured with an inexpensive memory, and the cost can be reduced. Further, the distortion is calculated from the pixel data of the search window and the pixel data of one current image block, and the pixel data of the search window discharged from the first and second pixel data transfer holding means at this time is converted into the third pixel. Since the pixel data in the search window held by the data transfer holding means can be returned to the first and second pixel data transfer means again to calculate the distortions corresponding to different current image blocks, the high speed transfer storage can be performed. The amount of data for accessing the desired pixel data of the search window to the means can be reduced, and the access time of the pixel data of the search window can be relatively shortened. Therefore, the motion vector search process can be speeded up.

According to the invention of claim 9, in the invention of claim 8, the pixel data of another search window to be processed next is read from the reference image storage means in advance and stored in the high-speed transfer storage means. Can be kept,
Regarding the time for supplying the pixel data of the search window to the first and second pixel data transfer holding means, the time for transferring the pixel data from the reference image storage means to the high-speed transfer storage means is ignored, and the transfer from the high-speed transfer storage means is performed. Only time is available. Therefore, the motion vector search process can be speeded up.

According to the invention of claim 10, claim 4
In the invention described in any one of (1) to (9), the pixel data of each pixel column of the search window held in the third pixel data transfer holding means can be returned together in all rows in the row direction opposite to the transfer path. Therefore, it is possible to easily and quickly return to the position where the distortion calculation was started.

According to the invention of claim 11, claim 2
In any of the inventions 5 to 8 and 10 to 10, when the current image block has an even number of pixel columns, when the distortion calculation is completed, the pixel data of the Mth column of the search window are respectively at the start of the distortion calculation. Since it is held in the first and second registers in the first column of the same row, the pixel data of the Mth column of the search window should be moved only in the opposite row direction without moving in the column direction of the transfer path. It is possible to return to the position at the beginning of distortion calculation simply by transferring. Therefore, the pixel data of the Mth column of the search window can be easily and quickly returned to the position at the time of starting the distortion calculation.

According to the invention of claim 12, claim 2
In the invention described in any one of 3 and 6 to 10, when the current image block has an odd number of pixel columns, when the distortion calculation is completed, (N-1) rows are shifted in the column direction from the beginning of the distortion calculation. Since the pixel data of the Mth column of the search window held in the first and second registers of the first column can be held in the third register, the pixel data of the Mth column of the search window can be stored in the column of the transfer path. It is possible to return to the first and second registers at the time of the distortion calculation start only by transferring so as to move only in the opposite row direction without moving in the direction. Therefore, the pixel data of the Mth column of the search window can be easily and quickly returned to the position at the time of starting the distortion calculation.

According to the invention of claim 13, claim 4
In the invention described in any one of (1) to (12), it is possible to control each of the arithmetic units of the first register and the distortion calculation means so as to perform the same operation at the same time by a common control signal. It is possible to perform high-speed processing by. According to the invention described in claim 14, in the invention described in any one of claims 4 to 13, the same number of second registers are arranged in the processor elements in the first row and the (H-N + 1) th row of each column. Therefore, the circuit can be easily formed as a buffer area for transferring the pixel data of the search window so as to reciprocate in the column direction.

According to the invention of claim 15, claim 4
15. In the invention according to any one of claims 1 to 13, since the first register and the second register can be electrically connected in a ring shape for each column, a second half of the second register according to claim 14 is formed. The register can constitute the second pixel data transfer holding means. Further, since the distance between the registers connected in a ring shape for each column can be arranged uniformly, a short transfer bus can be formed between the registers and the transfer time between the registers can be made uniform. Can be Therefore, a stable circuit with few errors can be formed, and the circuit scale can be reduced.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a motion vector search device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a search range of a current image block and a search window applied to the motion vector search device according to the embodiment of the present invention.

FIG. 3 is a diagram showing a configuration of a motion vector search device according to a first embodiment of the present invention.

FIG. 4 is a diagram showing a current image block and a search window applied to the motion vector search device according to the first embodiment of the present invention.

5 is a diagram showing configurations of a window data transfer unit 3000 and a horizontal side register unit 5000. FIG.

FIG. 6 is a diagram showing a current image block and a search window applied to the motion vector search device according to the first embodiment of the present invention.

FIG. 7 is a diagram showing signals output from the system control unit 6000 to each unit.

FIG. 8 is a time chart showing operations of a search window transfer unit 3000 and a horizontal side register unit 5000.

FIG. 9 is a time chart showing operations of the search window transfer unit 3000 and the horizontal side register unit 5000.

FIG. 10 is a time chart showing operations of the search window transfer unit 3000 and the horizontal side register unit 5000.

FIG. 11 is a time chart showing operations of the search window transfer unit 3000 and the horizontal side register unit 5000.

FIG. 12 is a time chart showing operations of search window transfer unit 3000 and horizontal side register unit 5000.

FIG. 13 is a time chart showing the operation of the similar block specifying unit 4000.

FIG. 14 is a time chart showing the operation of the similar block specifying unit 4000.

FIG. 15 is a diagram showing a terminal arrangement of a processor element PE.

FIG. 16 is a block diagram of a processor element PE.

FIG. 17 is a diagram showing a terminal arrangement of the vertical side register VS and a block diagram of the vertical side register VS.

FIG. 18 is a diagram showing a terminal arrangement of an input register IR and a block diagram of the input register IR.

FIG. 19 is a block diagram of the horizontal side register HS showing the terminal arrangement of the horizontal side register HS.

20 is a block diagram of a similar block identifying unit 4000. FIG.

FIG. 21 is a block diagram of a flip-flop with selector 4400.

22 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the first clock of the clock pulse signal CK1. FIG.

23 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the second clock of the clock pulse signal CK1. FIG.

FIG. 24 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the third clock of the clock pulse signal CK1.

FIG. 25 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the fourth clock of the clock pulse signal CK1.

FIG. 26 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 6th clock of the clock pulse signal CK1.

27 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 10th clock of the clock pulse signal CK1. FIG.

28 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 11th clock of the clock pulse signal CK1. FIG.

29 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 16th clock of the clock pulse signal CK1. FIG.

30 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 17th clock of the clock pulse signal CK1. FIG.

31 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 19th clock of the clock pulse signal CK1. FIG.

32 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 20th clock of the clock pulse signal CK1. FIG.

33 is a diagram showing a transfer state of pixel data of a search window in the window data transfer unit 3000 and the horizontal side register 5000 at the 21st clock of the clock pulse signal CK1. FIG.

FIG. 34 is a diagram showing an operation of obtaining a motion vector MV by pipeline processing.

FIG. 35 is a diagram showing the configuration of a motion vector search device according to a second embodiment of the present invention.

FIG. 36 is a diagram showing the configuration of a window data transfer unit 3001 of the motion vector search device according to the third embodiment of the present invention.

FIG. 37 is a diagram in which each processor element and each vertical side register VS are connected in a ring shape for each column and arranged so that their intervals are equal to each other.

FIG. 38 is a diagram showing the configuration of a motion vector search device according to a fourth embodiment of the present invention.

[Fig. 39] Fig. 39 is a diagram illustrating a simple interframe predictive coding method.

[Fig. 40] Fig. 40 is a diagram illustrating a motion compensation interframe predictive coding method.

41 is a diagram showing a current image block 220 on a current image 110 and a search window 410 and a candidate block 302 on a reference image 310. FIG.

FIG. 42 is a current image block 220, search window 4
It is a figure which shows the relationship between 10 and the candidate block 302.

43 is a diagram showing a positional correspondence relationship between pixel data in the current image block 220 and pixel data in each candidate block 302. FIG.

FIG. 44 is a diagram showing an operation of calculating each distortion in each processor element based on the pixel data of the current image block and the pixel data of each candidate block.

FIG. 45 is a diagram showing an operation of calculating each distortion in each processor element based on the pixel data of the current image block and the pixel data of each candidate block.

[Explanation of symbols]

1 First Pixel Data Transfer Holding Means 2 Second Pixel Data Transfer Holding Means 3 Third Pixel Data Transfer Holding Means 4 Window Data Transfer Control Means 5 Distortion Calculating Means 6 Similar Block Identifying Means 7 Ejection Data Hold Control Means 8 Return Data Transfer Controls Means 9 High-speed transfer storage means 10 Human image 20,21 Significant pixel area 100,110 Current image 201,202,203,211,212,213,2
14,220 Current image block 300,310 Reference image 301,302 Candidate block 401,402,403,410 Search window 501,502,503,504,505,511,5
12,513 Search range 1000 Current image memory 1001 Current image cache memory 2000 Reference image memory 2001 Reference image cache memory 3000,3001,3002 Window data transfer unit 3100 Processor unit 3200 First vertical side register unit 3300 Second vertical side register unit 3400 Input register unit 3600 Transfer direction switching unit 3610, 3611, 3612, 3613 Selector 3620, 3621, 3622, 3623 Flip-flop 3700 Distortion calculation unit 3710 Subtractor 3720 Positive number converter 3730 Adder 3740 Flip-flop 3750 Inverter 3760 Logical product operation 3800 Distortion transfer unit 3810 Selector 382 Flip-flop 4000 Similar block identification unit 4100 Minimum distortion detection unit 4101 Comparator 4102 Logical sum operator 4103 Comparator 4104 Selector 4105 Flip-flop 4106 Selector flip-flop 4200 Motion vector vertical component detection unit 4201 Selector 4202 Flip-flop 4203 Conversion table 4204 Selector Flip-flop 4300 Motion vector horizontal component detection unit 4301 Counter 4302 Selector 4303 Flip-flop 4304 Conversion table 4305 Flip-flop with selector 4400 Flip-flop with selector 4401 Selector 4402 Flip-flop 5000 Horizontal side register unit 6000 System control unit Tsu door

Claims (15)

[Claims] 1. A method of searching for a motion vector used to predict a current image partially forming a moving image based on a reference image partially forming the moving image, wherein the current image is a pixel. Each of the current image blocks includes a plurality of current image blocks represented by pixels in a plurality of rows and a plurality of columns having data, and the reference image includes a plurality of candidate blocks having the same size as the current image block having pixel data. A motion vector search method for specifying the motion vector based on a block position of any one candidate block similar to the current image block among the plurality of candidate blocks and a block position of the current image block, Pixel data of a plurality of current image blocks and at least two candidate blocks of the plurality of candidate blocks of the reference image And a pixel area of each candidate block in the search window is formed in the search area and a search area corresponding to the size of the search window and the current image block is formed. A pixel is provided between the first pixel data transfer holding means for inputting and holding a part of the data and transferring the held pixel data in a predetermined direction of the search area, and the first pixel data transfer holding means. Second pixel data transfer holding means for transferring data and transferring the pixel data of the search window along with the first pixel data transfer holding means along a predetermined transfer path passing through the search area; and along the transfer path Third pixel data transfer holding means for inputting and outputting a part of the pixel data transferred by the above-mentioned first and second pixel data transfer holding means And, each of the pixel data of the search window is input to the first and second pixel data transfer / holding means, and the pixel data of each pixel column of the search window is arranged in all rows together in the column direction. A window data transfer step of transferring the pixel data of the search window along the transfer path by the first and second pixel data transfer holding means so as to move in the row direction while reciprocating to and from the current image. Based on the pixel data of the block and the pixel data held in the first pixel data transfer holding means,
For each of the current image blocks, a first distortion calculating step of calculating a plurality of distortions representing image differences between the current image block and the plurality of candidate blocks, and for each of the current image blocks. A similar block specifying step of detecting the minimum value of the plurality of calculated distortion values and specifying the similar one candidate block, and further, the first and second pixel data transfer holding Of the pixel data of the search window inputted to the means, the pixel data of each column from the first column of the search window by a predetermined number of columns are discharged from the first and second pixel data transfer holding means and the third And a step of holding the discharge data held by the pixel data transfer holding means, and the image held by the third pixel data transfer holding means. Raw data is transferred from the third pixel data transfer holding means to the first pixel data transfer holding means.
And a pixel data returning step of returning to the second pixel data transfer holding means, and the current image block using the pixel data returned to the first and second pixel data transfer holding means in the pixel data returning step. A second distortion calculation step of performing distortion calculation corresponding to different current image blocks, and a motion vector search method. 2. An apparatus for searching for a motion vector used for predicting a current image partially forming a moving image based on a reference image partially forming the moving image, wherein the current image is a pixel. Each of the current image blocks includes a plurality of current image blocks represented by pixels in a plurality of rows and a plurality of columns having data, and the reference image includes a plurality of candidate blocks having the same size as the current image block having pixel data, respectively. A motion vector search device that specifies the motion vector based on a block position of any one candidate block similar to the current image block among the plurality of candidate blocks and a block position of the current image block, Pixel data of a plurality of search windows, each of which stores pixel data of a reference image and includes at least two candidate blocks. A reference image data storage unit capable of outputting and a search area corresponding to the size of the search window and the current image block are formed, and a part of pixel data of each candidate block in the search window is input into the search area. While holding the pixel data, the pixel data is transferred between the first pixel data transfer holding means for transferring the held pixel data in a predetermined direction of the search area, and the first pixel data transfer holding means, Second pixel data transfer holding means for transferring the pixel data of the search window along the predetermined transfer path passing through the search area together with the first pixel data transfer holding means, and transferred along the transfer path. Third pixel data transfer holding means for inputting and outputting a part of the pixel data from the first and second pixel data transfer holding means, and the search window Window pixel data is input to the first and second pixel data transfer / holding means, at least two candidate blocks of the plurality are sequentially input into the search area, and pixels of each pixel column of the search window are input. Window data for causing the pixel data of the search window to be transferred along the transfer path by the first and second pixel data transfer holding means so that the data moves in the row direction while reciprocating in all columns together in the column direction. Transfer control means, based on the pixel data of the current image block and the pixel data held in the first pixel data transfer holding means,
For each of the current image blocks, distortion calculation means for respectively calculating a plurality of distortions that represent image differences between the current image block and the plurality of candidate blocks, and for each of the current image blocks. Pixel data is transferred by the similar block specifying unit that detects the minimum value of the plurality of distortion values and specifies the similar one candidate block, and the first and second pixel data transfer holding units. At this time, among the pixel data input to the first and second pixel data transfer holding means, the pixel data of each column from the first column of the search window by a predetermined column are transferred to the first and second pixel data. Discharge data holding control means for causing the third pixel data transfer holding means to hold the discharged data while discharging from the holding means; Return data transfer control means for returning the pixel data held in the third pixel data transfer holding means by the holding control means from the third pixel data transfer holding means to the first and second pixel data transfer means, The distortion calculation means uses the pixel data returned from the third pixel data transfer holding means to the first and second pixel data transfer holding means by the return data transfer control means, and A motion vector search device characterized by performing distortion calculation corresponding to a current image block different from an image block. 3. The return data transfer control means causes the third pixel data transfer holding means to return the pixel data of the search window to the state at the time of starting the distortion calculation after the distortion calculation is completed by the distortion calculating means. While returning the held pixel data of the search window to the first and second pixel data transfer holding means, the window data transfer control means
The pixel data of the search window held by the first and second pixel data transfer holding means is transferred in the direction opposite to the transfer path, and the distortion calculation means changes the current image block different from the current image block. The motion vector search device according to claim 2, wherein a corresponding distortion calculation is performed. 4. An apparatus for searching for a motion vector used for predicting a current image partially forming a moving image based on a reference image partially forming the moving image, wherein N,
When M is an integer, the current image includes a plurality of current image blocks represented by N rows and M columns of pixels having pixel data, and the reference image has the same size as the current image block having pixel data. Any one of the plurality of candidate blocks, which is similar to the current image block among the plurality of candidate blocks,
A motion vector search device that specifies the motion vector according to the block position of one candidate block and the block position of the current image block, and stores pixel data of the reference image when H and L are integers. A reference image data storage unit capable of outputting pixel data of a plurality of search windows of H rows and L columns each including at least two candidate blocks, and (H-N + 1) according to the sizes of the search window and the current image block. ) Forming a search area of row (L−M + 1) columns, inputting and holding a part of pixel data of each candidate block in the search window in the search area, and holding the held pixel data in the search area. (H−N + 1) × (L−M + 1) first 1
A first pixel data transfer holding means having a register, and a plurality of second registers for transferring pixel data between the first register of the first pixel data transfer holding means,
Second pixel data transfer holding means for transferring the pixel data of the search window along the predetermined transfer path passing through the search area together with the first pixel data transfer holding means, and transferred along the transfer path. A third having (M−1) × H third registers for inputting and outputting a part of the pixel data from the first and second pixel data transfer holding means.
And pixel data of the search window is input to the first and second pixel data transfer and holding means, and at least two candidate blocks of the plurality are sequentially input into the search area. , So that the pixel data of each pixel column of the search window moves in the row direction while reciprocating in the column direction together with all the rows, the pixel data of the search window is transferred by the first and second pixel data transfer holding means. Based on window data transfer control means for transferring along the transfer path, and pixel data of the current image block and pixel data held in the first pixel data transfer holding means,
For each of the current image blocks, distortion calculation means for respectively calculating a plurality of distortions that represent image differences between the current image block and the plurality of candidate blocks, and for each of the current image blocks. When the minimum value of the plurality of distortion values is detected and the similar block specifying unit that specifies the similar one candidate block and the pixel data are transferred by the first and second pixel data transfer holding units Of the pixel data input to the first and second pixel data transfer holding means, the pixel data of each column from the first column to the (M-1) th column of the search window are set to the first and second pixels. Discharge data holding control means for holding the third pixel data transfer holding means while discharging the data from the data transfer holding means; Return data transfer control means for returning the pixel data held in the third pixel data transfer holding means by the holding control means from the third pixel data transfer holding means to the first and second pixel data transfer means, The distortion calculation means uses the pixel data returned from the third pixel data transfer holding means to the first and second pixel data transfer holding means by the return data transfer control means, and A motion vector search device characterized by performing distortion calculation corresponding to a current image block different from an image block. 5. The return data transfer control means returns the pixel data of the (M-1) th column from the third register to the first and second registers, and at the same time, the pixel of the first column of the search area. The pixel data held in the first and second registers corresponding to the columns are transferred so as to be returned to the other side in the row direction, and the pixel data for M columns corresponds to any of the current image blocks. When returning to the position at the time of starting the distortion calculation, the transfer direction of the input pixel data is switched again to the one side in the row direction, and the distortion calculating means is returned from the third register to the first and second registers. When the pixel data for (M-1) columns and the pixel data for (M-1) columns are input to the third register, a first pixel column corresponding to the first pixel column in the search area is input. 5. The motion vector search device according to claim 4, wherein distortion calculation corresponding to a current image block different from the current image block is started using the pixel data held in the second register and the pixel data held in the second register. 6. An apparatus for searching for a motion vector used to predict a current image partially forming a moving image based on a reference image partially forming the moving image, wherein N,
When M is an integer, the current image includes a plurality of current image blocks represented by N rows and M columns of pixels having pixel data, and the reference image has the same size as the current image block having pixel data. Any one of the plurality of candidate blocks, which is similar to the current image block among the plurality of candidate blocks,
A motion vector search device for specifying the motion vector according to a block position of one candidate block and a block position of the current image block, which stores pixel data of the reference image and includes at least two candidate blocks each. Reference image data storage means capable of outputting pixel data of a plurality of search windows of H rows and L columns, and search of (H-N + 1) rows (L-M + 1) columns according to sizes of the search window and the current image block. A region is formed, a part of the pixel data of each candidate block in the search window is input and held in the search region, and the held pixel data is transferred in a predetermined direction of the search region (H-N + 1). ) × (L−M + 1) first
A first pixel data transfer holding means having a register, and a plurality of second registers for transferring pixel data between the first register of the first pixel data transfer holding means,
Second pixel data transfer holding means for transferring the pixel data of the search window along the predetermined transfer path passing through the search area together with the first pixel data transfer holding means, and transferred along the transfer path. Third pixel data transfer holding means having M × H third registers for inputting and outputting a part of the pixel data from the first and second pixel data transfer holding means, and pixel data of the search window While inputting to the first and second pixel data transfer holding means, at least two candidate blocks of the plurality are sequentially input into the search area, and the pixel data of each pixel column of the search window is stored in all rows. The first and second parts are arranged so as to move in the row direction while reciprocating in the column direction together.
Held by the window data transfer control means for transferring the pixel data of the search window along the transfer path by the pixel data transfer holding means, and the pixel data of the current image block and the first pixel data transfer holding means. Based on pixel data and
For each of the current image blocks, distortion calculation means for respectively calculating a plurality of distortions that represent image differences between the current image block and the plurality of candidate blocks, and for each of the current image blocks. When the minimum value of the plurality of distortion values is detected and the similar block specifying unit that specifies the similar one candidate block and the pixel data are transferred by the first and second pixel data transfer holding units Of the pixel data input to the first and second pixel data transfer holding means, the pixel data of each column from the first column to the Mth column of the search window is transferred to the first and second pixel data transfer holding means. Discharging data holding control means for holding the third pixel data transfer holding means while discharging the discharging data, and the discharging data holding system. Return data transfer control means for returning the pixel data held in the third pixel data transfer holding means by the means from the third pixel data transfer holding means to the first and second pixel data transfer means. The distortion calculation means uses the pixel data returned from the third pixel data transfer holding means to the first and second pixel data transfer holding means by the return data transfer control means to use the current image block. A motion vector search device characterized by performing distortion calculation corresponding to a current image block different from. 7. The return data transfer control means starts distortion calculation for pixel data of M columns from the 1st column to the Mth column of the search window from the third register to correspond to any one of the current image blocks. When returning to the position of time, the transfer direction of the input pixel data is switched again to the one side in the row direction, and the distortion calculating means is equivalent to the M columns returned from the third register to the first and second registers. 7. The motion vector search device according to claim 6, wherein distortion calculation corresponding to a current image block different from the current image block is started using the pixel data of. 8. An apparatus for searching for a motion vector used to predict a current image partially forming a moving image based on a reference image partially forming the moving image, wherein the current image is a pixel. Each of the current image blocks includes a plurality of current image blocks represented by pixels in a plurality of rows and a plurality of columns having data, and the reference image includes a plurality of candidate blocks having the same size as the current image block having pixel data, respectively. A motion vector search device that specifies the motion vector based on a block position of any one candidate block similar to the current image block among the plurality of candidate blocks and a block position of the current image block, Pixel data of a plurality of search windows, each of which stores pixel data of a reference image and includes at least two candidate blocks. A reference image data storage unit capable of outputting and a search area corresponding to the size of the search window and the current image block are formed, and a part of pixel data of each candidate block in the search window is input into the search area. While holding the pixel data, the pixel data is transferred between the first pixel data transfer holding means for transferring the held pixel data in a predetermined direction of the search area, and the first pixel data transfer holding means, Second pixel data transfer holding means for transferring the pixel data of the search window along the predetermined transfer path passing through the search area together with the first pixel data transfer holding means, and transferred along the transfer path. Third pixel data transfer holding means for inputting and outputting a part of the pixel data from the first and second pixel data transfer holding means, and the reference image data The pixel data is read from the data storage unit for each predetermined pixel area including a part of the pixel data of any one of the plurality of search windows and is stored, and
High-speed transfer storage means for supplying the stored pixel data of the search window to the first and second pixel data transfer holding means at a transfer rate higher than the data transfer rate from the reference image data storage means; The pixel data of the search window stored in the means is input to the first and second pixel data transfer holding means, and at least two candidate blocks of the plurality are sequentially input into the search area, and
The pixel data of the search window is transferred by the first and second pixel data transfer holding means so that the data of each pixel column of the search window is moved in the row direction while reciprocating in all the rows together in the column direction. Based on window data transfer control means for transferring along a path, pixel data of the current image block and pixel data held in the first pixel data transfer holding means,
For each of the current image blocks, distortion calculation means for respectively calculating a plurality of distortions that represent image differences between the current image block and the plurality of candidate blocks, and for each of the current image blocks. When the minimum value of the plurality of distortion values is detected and the similar block specifying unit that specifies the similar one candidate block and the pixel data are transferred by the first and second pixel data transfer holding units Of the pixel data input to the first and second pixel data transfer holding means, pixel data of each column from the first column of the search window by a predetermined number of columns are transferred from the first and second pixel data transfer holding means. Discharge data holding control means for holding the third pixel data transfer holding means while discharging, and the discharge data holding system Return data transfer control means for returning the pixel data held in the third pixel data transfer holding means by the means from the third pixel data transfer holding means to the first and second pixel data transfer means. Among the pixel data for the predetermined pixel area read and stored from the reference image data storage means, the high-speed transfer storage means stores pixel data in a transfer order subsequent to the pixel data returned in the search area. The first and second pixel data transfer holding means are input again, and the distortion calculating means is controlled by the return data transfer controlling means from the third pixel data transfer holding means to the first and second pixel data transfer holding means. The pixel data returned to the first and second pixel data transfer holding means from the high-speed transfer storage means Using the chromatography data, the motion vector search apparatus characterized by performing the distortion calculation corresponding to different current picture block from the current picture block. 9. Pixel data is returned from said third pixel data transfer holding means to said first and second pixel data transfer holding means, and pixel data in a transfer order following said pixel data is transferred from said high speed transfer storage means. When part of the pixel data of any one of the plurality of search windows is supplied to the search area by being input again to the first and second pixel data transfer holding means, any one of the search 9. The motion vector search device according to claim 8, wherein the pixel data of the remaining part of the window is read from the reference image data storage means and stored in the high-speed transfer storage means. 10. The third pixel data transfer holding means comprises:
10. The motion vector search device according to claim 4, wherein the pixel data is input and output while maintaining the position of the pixel data at the start of the distortion calculation in the search area. 11. The motion vector search device according to claim 2, wherein each of the current image block and the candidate block has an even number of pixel columns. 12. The motion vector search device according to claim 2, wherein each of the current image block and the candidate block has an odd number of pixel columns. 13. The distortion calculating means is composed of an arithmetic circuit having a systolic array structure having a plurality of arithmetic units arranged two-dimensionally together with the first register. The motion vector search device according to any one of 1. 14. Among the plurality of second registers, some of the second registers have the same number in each column (N−1) × (L−).
M + 1) pieces of pixel data are provided, and the pixel data transferred in one direction in the column direction of the search area is input and output, and
(N-1) x so that the remaining second registers have the same number in each column
14. The motion vector search device according to claim 4, wherein (L−M + 1) pieces of pixel data are provided, and the pixel data transferred in the other direction of the column of the search area is input and output. . 15. The (N-1) × (L−M + 1) number of the second registers are provided so that each column has the same number, and the second registers are interposed between the first registers corresponding to the pixels at both ends in the column direction of the search area. The pixel data is transferred by performing the above process.
13. The motion vector search device according to any one of 13.