JPH08223578A - Method for searching motion vector and device therefor - Google Patents

Abstract

PURPOSE: To reduce the operation amount while searching without lowering the detection accuracy of a motion vector in a block matching type motion vector detection. CONSTITUTION: A small area division part 9 reads the image 6 of the present frame, divides the image into plural blocks, calculates the feature amount of each block and integrates the blocks where the differences of the feature amount is within threshold into one block. A small area division part 10 reads the image 8 of the reference frame conforming to the shape of the integrated block of the present frame. Thinning processing parts 21 and 22 perform the thinnings of the picture elements of a present frame block signal 14 and a reference frame block signal 17. A matching error calculation part 27 calculates the difference value 28 of the block signals 23 and 24. When the difference value 28 is smaller than the value of a memory 30, a comparator 29 stores the value 31 that a present frame block reading start address signal 16 is subtracted from a reference frame block reading start address signal 18 in a vector memory 32.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding system, and more particularly to a motion vector search method and apparatus for dividing a frame signal of an image into a plurality of blocks and obtaining motion vectors of the plurality of blocks.

[0002]

2. Description of the Related Art In a block matching type motion vector search in which an image is divided into rectangular blocks and a motion vector is searched, a full search method is used as a method capable of obtaining the highest accuracy. This is to set the search range centered on the block for which the motion vector is to be calculated in the referenced image, and to set the block that is most similar to the target block within the search range to the absolute value difference or the sum of squared differences. This is a method of finding out by using, etc. as an evaluation function and using the relative position as a motion vector. In the past, each block independently obtained a motion vector.

[0003]

In the full search method, since the evaluation function is calculated in the entire search range, the amount of calculation becomes enormous. As a method to easily reduce the calculation amount, it is possible to thin out the pixels in the block when evaluating the absolute value difference or the sum of squared differences, but the number of sample points at the time of matching is reduced, so false detection is performed. There was a problem that the possibility of

An object of the present invention is to solve the above problems and provide a motion vector search method and apparatus for reducing the amount of calculation at the time of search without lowering the accuracy of motion vector detection.

[0005]

A first motion vector search method of the present invention divides a frame signal of an image into a plurality of blocks of k pixels × j lines (k and j are integers of 1 or more), The feature amount of a block is calculated, and the difference between the feature amounts of the blocks is integrated into one block within a predetermined threshold value, and the blocks are combined according to the size of the integrated block. Pixels inside are thinned out, and block matching is performed using the integrated block in which pixels are thinned out to obtain a motion vector of the combined block.

The second motion vector search method of the present invention is
The frame signal of the image is divided into a plurality of blocks of k pixels × j lines (k and j are integers of 1 or more), the feature amount of each block is calculated, and the difference between the feature amounts of the blocks is predetermined. The blocks within the threshold are integrated into one block, pixels in the integrated block are thinned according to the size of the integrated block, and block matching is performed using the combined block in which the pixels are thinned. , Obtaining a motion vector of the integrated block, returning the integrated block to a plurality of blocks of k pixels × j lines, and setting the previously obtained motion vector as an initial transition,
Block matching is performed in a minute area around the motion vector to obtain a motion vector corresponding to a plurality of blocks of k pixels × j lines.

The first motion vector search device of the present invention is
A first frame memory for storing the image of the current frame,
A second frame memory for storing the image of the reference frame,
A vector memory for storing a motion vector and an image of the current frame stored in the first frame memory are read out, and the read out image is k pixels × j lines (k, j
Is a whole number of 1 or more), the feature amount of each block is calculated, and the difference between the feature amounts of the blocks is integrated into one block when the difference is within a predetermined threshold value. Then, a block signal of the integrated current frame, a read start address of the current frame block signal, and a first small area division unit that outputs integrated block information indicating the integrated block, and the second frame memory. Read out the image of the reference frame stored in
A second small area dividing section which reads out an image signal suitable for the shape of the integrated block based on the integrated block information, outputs the read image signal as a reference frame block signal, and outputs a read start address thereof; A current frame block signal and the integrated block information are input, the number of pixels in a block is counted from the integrated block information, pixels are thinned according to the number of pixels, and a block signal in which pixels are thinned is output. 1, a reference frame block signal and the integrated block information are input, the number of pixels in a block is calculated from the integrated block information, pixels are thinned according to the number of pixels, and pixels are thinned. With respect to each block signal output from the second thinning processing unit that outputs the thinned block signal and the first thinning processing unit, There is a matching error calculation unit that calculates the difference value of pixel values between a plurality of block signals within the search range output from the decimation processing unit, adds these, and outputs the difference value, and a minimum value memory. If the difference value output from the matching error calculation unit is smaller than the value stored in the minimum value memory, the value in the minimum value memory is updated with the difference value, and the read start address of the reference frame block is updated. And a comparator for writing the obtained address to an address corresponding to the block indicated by the integrated block information of the vector memory, by obtaining the difference between the read start address of the current frame block.

The first motion vector search device of the present invention is
A first frame memory for storing the image of the current frame,
A second frame memory for storing the image of the reference frame,
A vector memory that stores a motion vector and an image of the current frame stored in the first frame memory at the time of the first motion vector search are read, and the read image is k pixels × j lines (k and j are 1 (Integer above) divided into rectangular blocks, the feature amount of each block is calculated, and the difference between the feature amounts of the blocks is integrated into one block within a predetermined threshold value. An integrated current frame block signal, a read start address of the block signal of the current frame, a first small area division unit that outputs integrated block information indicating an integrated block, and a second motion vector search, The image of the current frame stored in the first frame memory is read, and the read image is read by k pixels × j lines (k, j
Is an integer greater than or equal to 1), and is divided into rectangular blocks, and a current frame block signal and a second small area division unit that outputs a read start address of the current frame block signal
At the time of the first motion vector search, the image of the reference frame stored in the second frame memory is read, the image signal that matches the shape of the integrated block is read by the integrated block information, and the read image The signal is used as a reference frame block signal, the read start address is output, and when the second motion vector search is performed,
Similar to the first motion vector search, the image of the reference frame stored in the second frame memory is read by adding the read start address signal to the address stored in the vector memory, and the read image is read. The signal is used as a reference frame block signal, the third small area division unit that outputs the read start address, the current frame block signal that is output during the first motion vector search, and the integrated block information are input, and A first thinning processing unit that counts the number of pixels in the block from the integrated block information, thins the pixels according to the number of pixels, and outputs a block signal in which the pixels are thinned, and the first and second motion vectors. At the time of searching, the reference frame block signal and the integrated block information are input, and whether the integrated block information is A second thinning processing unit that calculates the number of pixels in the block, thins pixels according to the number of pixels, and outputs a block signal in which the pixels are thinned, and a first thinning process during the first motion vector search. For each block signal output from the processing unit, a difference value of pixel values between a plurality of block signals within the search range output from the second thinning processing unit is calculated, and these are added to obtain the first value. It is output as a difference value of 1, and when the second motion vector search is performed,
For the current frame block signal output from the small area dividing unit of, the difference value of pixel values is calculated between the plurality of block signals within the search range output from the third thinning processing unit, and these are calculated. A matching error calculation unit for adding and outputting as a second difference value, and a minimum value memory,
Is smaller than the value stored in the minimum value memory, the value in the minimum value memory is updated with the first difference value, and the read address signal of the reference frame block and the first small area division are updated. The difference between the read start address signals of the current frame block output from the unit is calculated, the obtained address is written to the address corresponding to the block indicated by the integrated block information of the vector memory, and the second difference value The value of the minimum value memory is updated to obtain the difference between the read address signal of the reference frame block and the read start address signal of the current frame block output from the second small area division unit,
It has a comparator which writes the obtained address in the address according to the block which the said integrated block information of the said vector memory shows.

[0009]

According to the first and third aspects of the present invention, a plurality of blocks are adaptively integrated during search, that is, blocks having similar variances and averages are integrated to form an integrated block. It is characterized in that block matching is performed after thinning out pixels in accordance with.

If pixels are thinned out when performing matching, the amount of calculation can be reduced, but the number of sample points decreases, which may lead to a decrease in detection accuracy. Therefore, if blocks that include the same motion are collectively detected as one block, the total number of sample points is increased, so that even if pixels are thinned out, deterioration of detection accuracy can be suppressed.

The motion vector search method of claim 1 is shown in FIGS.
This will be described with reference to FIG. The pixel frame signal is, for example, as shown in FIG.
It is assumed that the blocks are divided into A to P as shown in FIG. First, the feature amount of each of the blocks A to P is calculated. Although the blocks A to P are adjacent blocks in the square area in this example, they can be arbitrarily selected. As the feature amount, any value such as variance or average of brightness values can be used. Next, set the threshold Th,
A block whose difference between the feature amount of a certain block and the feature amount of another certain block is within Th is determined to belong to the same object, and matching is performed as one block. If the feature amounts of the blocks A and B are F _A and F _B , respectively, when the formula (1) is satisfied, the blocks A and B are regarded as one block for matching.

| F _A −F _B | <Th (1) This determination is performed for all combinations of blocks A to P, and blocks determined to be included in the same object are integrated as one block and matched. To do. By these integration processes, for example, as shown in FIG.
If P is integrated into three blocks, block A + E
+ F is used for matching, and the detected motion vector is the motion vector of blocks A, E, F, and the motion vector detected by block B + C + D + G + H + K + L + O + P is block B, C, D, G, H, K, L,
For motion vectors of O and P, and block I + J + M +
The motion vector detected in N is used as a block I, J, M, N
Is the motion vector of.

The matching of the integrated blocks is performed after thinning out pixels according to the number of pixels in the integrated block.
As a method of thinning out, in the case of FIG. 2, A + E + F is a combination of 3 blocks, so if the pixels are thinned out at a rate of 2 pixels for every 3 pixels, the number of sample points at the time of matching is 1.
It will be the same as when blocking. Similarly B + C + D + G + H
+ K + L + O + P: 8 pixels out of 9 pixels, I + J + M
For + N, a method of thinning out at a rate of 3 out of 4 pixels can be considered. FIG. 3 shows an example in which three blocks are combined. White circles are thinned pixels.

However, according to the motion vector search method of the first aspect, when an object includes a motion such as a zoom or a rotation that deforms on a two-dimensional plane, even the same object has different motions depending on the position. May be shown. In this case, the motion vector detection in the integrated block is likely to cause an erroneous detection.

Therefore, in the second and fourth aspects of the invention, in the block matching type motion vector search, in order to suppress the deterioration of the detection accuracy due to thinning out the pixels in the block, blocks including the same motion are integrated, Pixels in the block are thinned according to the size, and after matching is performed in the integrated block, the integrated block is divided into each block again, and the motion vector obtained by the integrated block is used as the initial transition of each block in a small area again. Search for. With this method, the motion vector can be detected with a small amount of calculation without degrading the detection accuracy.

A motion vector search method according to claim 2 is shown in FIGS.
This will be described with reference to FIG. The feature amount of each of the blocks A to P is calculated, the blocks are integrated, and the integrated blocks A + E + F, B + C + D + G + H + K + L + O +.
Pixels in P, I + J + M + N are thinned out, the matching is performed, and the motion vector of the integrated block is obtained, which is the same as the motion vector search method of claim 1. In the motion vector search method according to the second aspect of the present invention, the detected motion vector is then used as the initial transition of each of the blocks A, E, and F. Similarly, B + C + D + G + H + K + L + O +
The motion vector detected in P is converted into blocks B, C, D,
At the initial transition of G, H, K, L, O, P, I + J + M + N
The motion vector detected in step S1 is used as the initial transition of blocks I, J, M, and N. Finally, in each of the blocks A to P, a final motion vector is obtained by performing a search in a minute area (for example, ± 1 pixel range) centering on the initial transition.

[0017]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 4 is a block diagram of the motion vector search device according to the first embodiment of the present invention, FIG. 5 is a block diagram of the small divider unit 9 in FIG. 3, and FIG. 6 is a small divider unit 10 in FIG.
It is a block diagram of.

The motion vector search apparatus of the present embodiment includes frame memories 3 and 4, small area division units 9 and 10, thinning processing units 21 and 22, a matching error calculation unit 27, and a comparator 29.
And a vector memory 32.

First, the motion vector search range 41 is input to the small area dividing unit 10. Any value can be used as the search range 41.

The image 1 of the current frame is the frame memory 2
Is stored in The image 6 of the current frame stored in the frame memory 2 is read with the value of the address signal 5 consisting of two coordinates (x, y) as an address value and input to the small area dividing unit 9. From the image 6 of the current frame, the small area dividing unit 9 calculates (k × m) × (j × n) (j, k, m, n is 1).
Block data of the above integer) is read (m = n = 4 in the example of FIG. 2), and the read data is further divided into k pixels × j.
After breaking it down into multiple rectangular blocks of lines,
The blocks are integrated by calculating the feature amount of each block and performing the comparison shown in Expression (1). The block signal 14 of the integrated current frame and the output block signal 1
The read start address 4 (address corresponding to the upper leftmost pixel of the block) 16 and integrated block information 15 indicating the integrated block are output. As shown in FIG. 5, the small area dividing unit 9 includes a block memory 33 and a feature amount calculating unit 3.
6 and the block integration determination unit 40. First, the image signal at the position corresponding to the upper left of the block signal to be read is read by giving the stored read start address to the frame memory 3 as the address signal 5. Thereafter, while changing the value of the address signal 5 so as to scan all the coordinates in the block to be read, the reading is performed (k × m) × (j × n) times and stored in the block memory 33. . Next, the feature amount calculation unit 36
Is stored in the block memory 33 by the same operation as reading from the block memory 33 (k × m) ×
The block signal of (j × n) is divided into k × j blocks by the value of the address signal 34 consisting of two coordinates of (x, y) and read out, and the feature amount is calculated for each read block. , Memory 37. The block integration determination unit 40 reads the feature amount 38 by the address signal 39, determines which block is integrated by the equation (1), and outputs integrated block information 15 indicating the integrated block. The block memory 33 outputs the integrated block signal 14 and the read start address 16 of the output block signal 14 according to the flag 15.

Next, the image 2 of the reference frame stored in the frame memory 4 is read out by using the value of the address signal 7 consisting of two components (x, y) as an address value and input to the small area dividing unit 10. It As shown in FIG. 6, the small area dividing unit 10 includes a block signal read control unit 42 and a counter 45. Block signal read control unit 42
The address signal 7 at the position corresponding to the upper left of the block signal to be read is given to the frame memory 4 to read the image signal 8 at that position, and at the same time, the value of the counter 44 is incremented by the count-up signal 43. At the same time, the address signal 18 at the upper left position is output to the comparator 29. Thereafter, while reading the block signal 17 while changing the value of the address signal 7 so as to scan all the coordinates in the block to be read until the value 44 of the counter 45 becomes k × j. The data is output to the thinning processing unit 22.

The small area dividing unit 10 uses the integrated block information 15
The image signal in the form of an integrated block is read out by the, and the read signal is output as the reference frame block signal 17, and at the same time, the read start address (x, y) at that time is output as the reference frame block read start address signal 18. To do.

The current frame block signal 14 and the reference frame block signal 17 are respectively thinned out by the thinning processing unit 2.
1, 22 are input. Each of the thinning processing units 21 and 22 calculates the number of pixels in the block from the integrated block information 15 and thins pixels according to the number of pixels,
The block signals 23 and 24 with the reduced number of pixels are output to the matching error calculation unit 27.

The matching error calculation unit 27 is composed of an arithmetic unit, calculates the difference square value of the pixel values of the block signals 23 and 24 output from the two decimation processing units 21 and 22, and adds them to obtain the difference. Comparator 2 with value 28
Output to 9. Further, in the matching error calculation unit 27,
The above calculation is performed with a plurality of reference frame block signals 24 within the search range corresponding to one current frame block signal 23. The difference value calculation is not limited to the difference squared, and the difference absolute value or the like can be used.

The comparator 29 has a minimum value memory 30, and when the input difference value 28 is smaller than the value stored in the minimum value memory 30, the value in the minimum value memory 23 is updated, and each of the two components is updated. A value obtained by subtracting the current frame block read start address signal 16 from the reference frame block read start address signal 18 composed of (x, y) or the current frame block read start address signal 1
The address 31 which is a value obtained by subtracting the reference frame block read start address signal 18 from 6 is used as the vector memory 3
Store in 2.

The vector memory 32 writes the address (motion vector) 31 output from the comparator 29 to an appropriate address in the memory according to the integrated block information 15. For example, in the integrated state as shown in FIG. 2, the motion vector calculated in the block A + E + F is
It is written in three addresses corresponding to E and F.

In the above operation, the reference block first scans the entire search range input to the small area dividing unit 10 (for example, if the search range is ± 2 both in horizontal and vertical directions, a rectangular area of 5 × 5 = 25 pixels). It is repeated until it finishes. At the time of the first operation, the value of the address signal 18 corresponding to the upper left address of the block is the coordinate value at the upper left of the search range. Hereinafter, the above operation is repeated for the entire search range while changing the value of the address signal 18 in the horizontal and vertical directions at each operation.

When the above operation for the search range is completed,
The detected motion vector is stored in the vector memory 32.

FIG. 7 is a block diagram of a motion vector search device according to the second embodiment of the present invention. Those having the same functions as those in FIG. 1 are designated by the same reference numerals.

The motion vector search apparatus of this embodiment is shown in FIG.
Of the small area dividing unit 13 and the changeover switches 25 and 26.
Is added, and the vector memory 32 is composed of two vector memories 32a and 32b.

First, the motion vector search range 41 at the time of the first search and at the time of the second search is input to the small area dividing unit 10. Any value can be used as the search range 41.

First, the first search is performed. The first search performs the same operation as that of the motion vector search device of the first embodiment (the changeover switches 25 and 26 are tilted to the upper side of the drawing, the small area division unit 13 is not operated), and is the output of the comparator 29. The address (motion vector) 31 is written to an appropriate address in the vector memory 32a according to the integrated block information 15. For example, in the integrated state as shown in FIG. 2, A +
The motion vector calculated in the E + F block is A,
It is written in three addresses corresponding to E and F.

After the above operation is completed within the motion vector search range in the reference frame, the vector memory 32a
The motion vector of the reference frame from the current frame when the integrated block is used is stored in. This value is used as an initial value in the second search.

Next, a second search is performed. The image 12 of the reference frame stored in the frame memory 3 is input to the small area dividing unit 13. The small area dividing unit 13 divides the image 12 of the current frame into k × j blocks and outputs a reference frame block signal 19 and a block read start address signal 20. Similarly, the image 8 of the reference frame is input again to the small area dividing unit 10, but the address signal 7 and the block read start address signal 18 are obtained by adding the (x, y) output value 33 from the vector memory 32a. Become.

The reference frame block signal 8 read out according to the address signal 7 is input to the small area dividing unit 10, divided into blocks of k pixels × j lines, and output as a reference frame block signal 17. At the same time, the block read start address signal 18 is output. Then, as in the case of the processing in the integrated block, the difference value is calculated in the matching error calculation unit 27, and the comparator 29
If the difference value input at is smaller than the value stored in the minimum value memory 30, the value in the minimum value memory 30 is updated, and the reference frame block read start address signal 18 consisting of two components (x, y) A value obtained by subtracting the frame block read start address signal 20 or the current frame block read start address signal 20
To reference frame block read start address signal 1
An address (motion vector) 31 which is a value obtained by subtracting 8 is stored in the vector memory 32b. During the second search, the changeover switches 25 and 26 in the matching error calculation unit 27 and the input unit of the comparator 29 are tilted downward.

When the above operation for the search range is completed,
The detected motion vector is stored in the vector memory 32b. Although the search range in the second search can be set arbitrarily, the motion vectors of the blocks included in the same object have almost the same value, so it is sufficient to search for a minute area.

The vector memories 32a and 32b are set to 1
In some cases, the motion vector obtained in the second search may be overwritten on the motion vector obtained in the first search.

[0039]

As described above, the present invention has the following effects. (1) According to the inventions of claims 1 and 3, a plurality of blocks are adaptively integrated at the time of search to form an integrated block, and pixels are thinned according to the size of the integrated block, and then block matching is performed. By performing the above, the motion vector can be detected with a small amount of calculation without degrading the detection accuracy. (2) According to the inventions of claims 2 and 4, matching is performed by an integrated block in which blocks including the same motion are integrated and the number of sample points is increased, and the result is used as an initial transition of each block to search again in a minute area. By doing so, the motion vector can be detected with a small amount of calculation without degrading the detection accuracy.

[Brief description of drawings]

FIG. 1 is a diagram showing an example in which a frame signal of an image is divided into a plurality of blocks.

FIG. 2 is a diagram showing an example in which blocks A to P of FIG. 1 are integrated.

FIG. 3 is a diagram showing an example in which pixels in integrated blocks A, E, and F of FIG. 2 are thinned out.

FIG. 4 is a block diagram of a motion vector search device according to the first embodiment of the present invention.

5 is a block diagram of a small area dividing unit 9 in FIG.

6 is a block diagram of a small area dividing unit 10 in FIG.

FIG. 7 is a block diagram of a motion vector search device according to a second embodiment of the present invention.

[Explanation of symbols]

1 image of current frame 2 image of reference frame 3 frame memory (current frame) 4 frame memory (reference frame) 5,11 read address signal of image signal of current frame 6,12 image of current frame 7 image of reference frame Read address signal 8 Image signal of reference frame 9,13 Small area dividing section (current frame) 10 Small area dividing section (reference frame) 14,19 Block signal of current frame 15 Integrated block information 16,20 Current frame block read start address 17 reference frame block signal 18 reference frame block read start address 21 thinning processing unit (current frame) 22 thinning processing unit (reference frame) 23 current frame block signal 24 reference frame block signal 25, 26 switching switch 27 Matching error calculation section 28 Difference value 29 Comparator 30 Minimum value memory 31 Address data (motion vector) 32 Vector memory 32a, 32b Vector memory 33 Output of vector memory 32a 34 Block read address signal 35 Block signal 36 Feature quantity Calculation unit 37 Memory 38 Block feature amount 39 Feature amount read address signal 40 Block integration determination unit 41 Motion vector search range 42 Block read signal control unit 43 Count up signal 44 Counter 45 value 45 Counter

Claims (4)

[Claims] 1. A frame signal of an image is divided into a plurality of rectangular blocks of k pixels × j lines (k and j are integers of 1 or more), a feature amount of each block is calculated, and a feature amount between each block is calculated. Difference is integrated into a block within a predetermined threshold value, pixels in the integrated block are thinned according to the size of the integrated block, and the integrated block in which pixels are thinned out A motion vector search method for obtaining a motion vector of an integrated block by performing block matching using. 2. A frame signal of an image is divided into a plurality of rectangular blocks of k pixels × j lines (k and j are integers of 1 or more), a feature amount of each block is calculated, and a feature amount between each block is calculated. Difference is integrated into a block within a predetermined threshold value, pixels in the integrated block are thinned according to the size of the integrated block, and the integrated block in which pixels are thinned out Block matching is performed to obtain a motion vector of the integrated block, the integrated block is returned to a plurality of blocks of k pixels × j lines, and the previously obtained motion vector is set as an initial transition, A motion vector search method in which block matching is performed in a minute area around a motion vector to obtain a motion vector corresponding to a plurality of blocks of k pixels × j lines. 3. A first frame memory for storing an image of a current frame, a second frame memory for storing an image of a reference frame, a vector memory for storing a motion vector, and a first memory for storing the vector in the first frame memory. The image of the current frame that is present is read, the read image is divided into rectangular blocks of k pixels x j lines (k and j are integers of 1 or more), the feature amount of each block is calculated, and the feature between each block is calculated. The block whose amount difference is within a predetermined threshold is integrated into one block, and the block signal of the integrated current frame, the read start address of the current frame block signal, and the integrated block are shown. A first small area dividing section that outputs integrated block information; and an image of a reference frame stored in the second frame memory, By the integrated block information,
A second small area division unit that reads out an image signal that matches the shape of the integrated block, outputs the read image signal as a reference frame block signal, and outputs the read start address, the current frame block signal, and the second frame region signal A first decimation processing unit for inputting integrated block information, counting the number of pixels in the block from the integrated block information, thinning pixels according to the number of pixels, and outputting a block signal with thinned pixels; The reference frame block signal and the integrated block information are input, the number of pixels in a block is calculated from the integrated block information, the pixels are thinned according to the number of pixels, and the block signal in which the pixels are thinned is output. The second decimation processing unit outputs the block signals output from the second decimation processing unit and the first decimation processing unit. Calculates the difference value of the pixel values between a plurality of blocks signals in the search range,
A matching error calculation unit for adding these and outputting as a difference value, and a minimum value memory, and if the difference value output from the matching error calculation unit is smaller than the value stored in the minimum value memory, The value of the minimum value memory is updated with a difference value, the difference between the read start address of the reference frame block and the read start address of the current frame block is obtained, and the obtained address is indicated by the integrated block information of the vector memory. A motion vector search device having a comparator for writing to an address corresponding to a block. 4. A first frame memory for storing an image of a current frame, a second frame memory for storing an image of a reference frame, a vector memory for storing a motion vector, and a first frame memory for searching a motion vector for the first time. Read the image of the current frame stored in the frame memory of 1,
The read image is divided into rectangular blocks of k pixels × j lines (k and j are integers of 1 or more), the feature amount of each block is calculated, and the difference in feature amount between the blocks is determined in advance. A block within a predetermined threshold value is integrated into one block, and a block signal of the integrated current frame, a read start address of the current frame block signal, and integrated block information indicating the integrated block are output. And a small area dividing unit for reading the image of the current frame stored in the first frame memory during the second motion vector search,
The read image is divided into rectangular blocks of k pixels × j lines (k and j are integers of 1 or more), and a second small area for outputting a current frame block signal and a read start address of the current frame block signal. Dividing unit and at the time of the first motion vector search, the image of the reference frame stored in the second frame memory is read, and the image signal suitable for the shape of the integrated block is read by the integrated block information. , The read image signal is used as a reference frame block signal, the read start address is output, and the second motion vector search is performed.
Similar to the first motion vector search, the image of the reference frame stored in the second frame memory is read by adding the read start address signal to the address stored in the vector memory, and the read image is read. A signal as a reference frame block signal, and a third small area dividing section for outputting a read start address thereof; the current frame block signal and the integrated block information output at the first motion vector search; Count the number of pixels in the block from the integrated block information,
A first decimation processing unit that decimates pixels according to the number of pixels and outputs a block signal in which the pixels are thinned, and the reference frame block signal and the integrated block information during the first and second motion vector searches. A second thinning processing unit that calculates the number of pixels in the block from the integrated block information, thins pixels according to the number of pixels, and outputs a block signal in which the pixels are thinned; At the time of motion vector search, for each block signal output from the first thinning processing unit, a difference value of pixel values between a plurality of block signals in the search range output from the second thinning processing unit is calculated. Calculate and add these together
Of the current frame block signal output from the second small area division unit during the second motion vector search. A matching error calculation unit that calculates a difference value of pixel values between the block signals and outputs the sum as a second difference value; and a minimum value memory, wherein the first difference value is the If it is smaller than the value stored in the minimum value memory, the value of the minimum value memory is updated with the first difference value, and the read start address signal of the reference frame block and the first small area dividing unit are output. Then, the difference between the read start address signals of the current frame block is obtained, the obtained address is stored in the vector memory, the value of the minimum value memory is updated with the second difference value, and the reference is made. A motion including a comparator for obtaining a difference between a read start address signal of a frame block and a read start address signal of the current frame block output from the second small area division unit, and storing the obtained address in the vector memory. Vector search device.