JPH01233893A - Moving vector detecting method - Google Patents

Abstract

PURPOSE:To save the number of times of evaluation calculation by generating a hierarchy picture from a picture signal so as to detect a rough moving vector and detecting moving vector sequentially with a fine accuracy. CONSTITUTION:Picture data GD1, GD2 of 1st and 2nd frames are stored respectively in 1st and 2nd hierarchy picture memories 2, 4, in hierarcy picture generating section 1, 3, the picture data of the 1st frame in the memory 2 and the picture data of the 2nd frame in the memory 4 are converted into the 1st-Kth hierarchy picture data whose resolution is sequentially made coarse and the result is stored in the memories 2, 4, and a moving vector detection section 5 reads the Kth hierarchy picture data with most rough resolution to obtain the moving vector, and then the (K-1)th hierarchy picture data is read and the moving vector is obtained based on the obtained moving vector. Then similar processing is repeated and then finally based on the data read from the 1st and 2nd frames, the moving vector is obtained based on the moving vector obtained form the 1st hierarchy picture data. Thus, the number of times of evaluation calculation is reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION "Field of Industrial Application" The present invention relates to a motion vector detection method used when transmitting moving images.

``Prior art'' In videophone calls, video conferences, etc., moving images are transmitted, and when transmitting these moving images, the number of bits of the image data to be transmitted must be reduced, or in other words, how can data be compressed? The important question is whether to do so.

Conventionally, an interframe predictive coding method is known as a method for compressing image data. This method uses the brightness of each pixel of the frame to be transmitted (for monochrome images)
This method calculates the difference between the image data representing the current frame and each image data of the previously transmitted frame, and transmits this difference data as the current frame display data instead of the image data. This method can perform significant data compression for images with no or little movement, as the correlation between frames is large, but for images with large movement, the correlation between frames is small, so the compression effect is low. There are few drawbacks.

A motion compensated interframe predictive coding method is known as a means to solve this problem. In this method, when obtaining inter-frame difference data, a motion vector is first obtained. This motion vector is a vector indicating the moving direction and moving distance of the image, and the image of the previous frame is, for example, the one shown by the solid line in FIG. When it moves, its motion vector becomes the vector V shown by the arrow in the figure. In practice, a frame is divided in advance into blocks each having, for example, 8×8 pixels, and a motion vector is determined for each block. Next, the image of the previous frame is moved according to each determined motion vector, and then difference data between the moved image and the image of the current frame is determined, and this difference data is transmitted.

According to this method, it is possible to obtain the same data compression effect even for moving images as for images with no movement.

Now, as a method for detecting the above-mentioned motion vector, (
1) There is a method in which evaluation values indicating the degree of suitability of all the motion vectors within the motion vector search range are obtained, and a motion vector that minimizes the evaluation value is obtained. However, this method requires a large number of times to calculate the evaluation value. For example, when the search range is ±7 pixels in both the vertical and horizontal directions, the number of evaluation value calculations is 225, which is not practical. On the other hand, there is a known method (2) in which a multi-stage tree search is performed on the original image when determining the evaluation value. According to this method, the number of evaluation value calculations can be significantly reduced. For example, in the case of the above-mentioned search range, a motion vector can be detected with 23 evaluation value calculations (see Japanese Patent Laid-Open No. 158784/1984).

"Problem to be solved by the invention" The present invention has been made in view of the above-mentioned circumstances.
It is an object of the present invention to provide a motion vector detection method that can further reduce the number of evaluation value calculations than the method 2).

"Means for Solving the Problems" FIG. 1 is a block diagram showing the principle of the invention. The present invention provides a motion vector detection method for detecting a motion vector without indicating the moving direction and moving distance of an image between a first frame and a second frame. and the image data GD2 of the second frame are stored in the first and second hierarchical image memories 2 and 2, respectively.
At the same time, the hierarchical image generation unit 1.3 converts the image data of the first frame in the memory 2 and the image data of the second frame in the memory 4 into first to fourth images whose resolutions are sequentially coarsened. The first step is to convert to hierarchical image data.
The motion vector detection section 5 first reads out the first hierarchical image data with the coarsest resolution from the memory 2.4 to obtain a motion vector. The hierarchical image data of (K-1) is read from the memories 2 and 4, and based on the read data, a motion vector is determined using the motion vector determined above as a reference. , said first
The image data of the frame and the second frame are read out, and based on the read data, a motion vector is determined based on the motion vector determined from the image data of the first layer, and the determined result is output. It is a feature.

Furthermore, in the above method, the motion vector obtained using coarse hierarchical image data of a higher resolution may be canceled out in processing based on fine hierarchical image data of a lower resolution.

Effect-1 Let us assume that the hierarchical image data is configured as shown in FIG. The 4 image data of the layer is averaged to form the image data of 1 pixel of the 2nd layer.In this case, the layer image is 1," because the total number of pixels decreases as the layer goes to the 2nd layer. If the aforementioned block size is fixed, the number of blocks will also decrease. For example, if the total number of pixels in the original frame is XXY pixels and the block size is N x M pixels, the number of blocks in the original image is (X/N) x (77M), as shown in Figure 2. When four pixels in the lower layer correspond to one pixel in the upper layer, the total number of pixels in the first layer image is (XX
Y)/4 pixels, the number of pixels is (X/N)
)/4 ([!!].Also, since one block in the upper layer has an influence on multiple blocks in the lower layer,
The motion vector detected in the top layer, that is, the second layer, is the first
This becomes the initial value for motion vector detection of the four blocks in the layer, and the motion vector determined in the first layer becomes the initial value for motion vector detection in the four blocks of the 0th layer, that is, the original frame.

Therefore, the motion vector detection work required to find the motion vector of one block in the original frame is 1 (original frame) + 1/4 (first layer) + l/16 (second layer) = 2
It only takes 1/16th of a time. In the case of this method, the number of evaluation value calculations in the search range of ±7 described in the conventional technique is 9×21/1.
6 = 11.8125 times, and the above methods (1) (2
) has been significantly reduced compared to - Numerically, assuming that the search range for each layer is R pixels, the number of layers is L layers, and the pixel reduction rate of the layer image is l/S in both the horizontal and vertical directions,
The motion vector search range is as follows, and the number of evaluation value calculations is as follows.

There are two ways to widen the search range: one is to widen the search range in each layer, and the other is to increase the number of layers. The former involves a significant increase in the number of calculations, whereas the latter, that is, according to the present invention, requires only a small amount of calculation. This only requires an increase in the number of calculations.

As described above, the motion vector detection method according to the present invention has the advantage of being able to detect motion vectors in a wide search range with a small number of evaluation value calculations.

"Embodiment" Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 3 is a block diagram showing the configuration of the motion vector detection circuit 11 according to an embodiment of the present invention. In this diagram,
12 is a terminal to which image data CDI of the current frame (the frame to be transmitted) is supplied; 13 is a terminal to which image data GD2 of the previous frame (the frame transmitted or reproduced one time before) is supplied; 14; .15
are hierarchical image data generation units, and 1.6.17 are hierarchical image memories, respectively.

The hierarchical image data generation section 14 is a circuit that generates the first layer image data and the second layer image data shown in FIG. 2 based on the current image data GD1, and as shown in FIG. Terminal 14t to which current image data GDl is supplied
, an adder circuit +4a, a latch 14b that latches the output of the adder circuit 14a, and a shifter 14Q that shifts the output of the latch 14b by 2 bits toward the lower bits, that is, reduces the output of the latch 1.4b to 1/4. , a control circuit 14d that controls each part, and an address generation circuit 14e that generates an address for the hierarchical image memory 16. Furthermore, the hierarchical image memory 16 has three storage areas EaO to Ea2, and is also provided with an input/output data terminal and an address terminal AD.

In such a configuration, the image data CDI (brightness variation data of each pixel) of the current frame is sequentially supplied to the terminal 14t, and this image detector GDI is connected to the input/output of the hierarchical image memory 16 via the bidirectional bus B1. The data is supplied to the data terminal and sequentially written into the storage area EaO of the memory 16. Then, all the image data of the current frame is stored in the storage area Ea.
After being written into the memory area EaO, the image data of the four pixels in the upper left section of FIG. It is considered to be 1/4 by
The result of this calculation, ie, the average value of the four image data, is written into the storage area Ea1 of the memory 16 as first layer image data.

Thereafter, four pieces of image data are sequentially read from the storage area EaO according to the classification shown in FIG. 5, an average value is calculated, and the calculation result is sequentially written to the storage area Eal. Note that the symbol GF in FIG. 5 indicates the current frame. When the above processing is completed for all the image data CDI of the current frame, the same processing as above is performed for each of the first layer image data in the storage area Eat. That is, first, four pieces of image data are sequentially read from the storage area Eal, sequentially accumulated by the adder circuit 14a and the latch 14b, and the accumulated result is 1/4 by the shifter 14c to obtain the average value of the four image data. This average value is written into storage area Ea2. Below, the above process is performed in area Eal.
Perform this for all image data within.

Note that in the above process, after all the image data is written in the storage area EaO, the data in the same area EaO is read out and processed. 1 ta or more) is in area Ea
The process for writing to the next area Eal may be started at the time the data is written to O. The same applies to the processing of data within area Eal.

In this way, the first layer image data and the second layer image data shown in FIG. 2 are generated, and the generated image data is written into the layered image memory 16. The hierarchical image data generation section 15 and the hierarchical image memory 17 are also similar to the hierarchical image data generation section 14 and the hierarchical image memory 16.
The image data GD2 of the previous frame is written to the storage area EbO of the hierarchical image memory 17, and is converted into the first layer image data and the second layer image data, and is stored in the storage areas Ebl and Eb2. Each is written.

Next, in FIG. 3, 20 is a hierarchical image memory 16, l
This is a motion vector detection unit that detects a motion vector based on the image data in 7. In this motion vector detection section 20, an address generation section 21 generates an address for scanning each pixel in an 8×8 pixel block that is a target of motion vector detection, and outputs the address as it is to the hierarchical image memory 16. On the other hand, the adder 2
2, the offset value is added and then output. The other input terminal of this adder 22 is connected to the initial motion vector latch 2.
3 and the output of the search vector generator 24 are added to the adder 2.
The addition result added in step 5 is supplied as an offset value. Note that the initial motion vector latch 23 and search vector generation section 24 will be explained later. The output latch 26 is a latch that reads the output of the adder 25 according to a command from the control unit 30, and the motion vector memory 27 is a memory in which the motion vector data read into the output latch 26 is stored.

The evaluation value calculation unit 28 calculates, for each block of image data sequentially read out from the hierarchical image memory 16.17,
An evaluation value is obtained by calculating the following equation (1) or (2), and the obtained evaluation value is output to the minimum value detection section 29.

-R(lx N+ i+ mx, JX N+ j+ m
y) l...(1) -R(lx N+ i+mx, Jx N+ j+ my
))'・・・・・・(2) Here, E(・) is the evaluation value, m is the horizontal and vertical components of the motion vector, and S(・) is the image data of the current frame or 1st layer image data or 2nd layer image data, R(・) is the image data of the previous frame or the 1st layer image data.
Hierarchical image data or second layer image data, I and J are data indicating the number from the left end and number from the top of the target block, respectively.N and M are the number of pixels in the horizontal and vertical directions of the block. be. Although the above formulas (1) and (2) are shown as general formulas, in this example, mX, my=-1, 0°1 N, M=8. The minimum value detection section 29 detects the minimum value among the evaluation values output from the evaluation value detection section 2B, and outputs data indicating a motion vector corresponding to this detection result to the control section 30.

Next, the operation of the motion vector detection section 20 with the above configuration will be explained. Note that the evaluation value detection unit 28 calculates the evaluation value based on equation (1).

Now, the screen GMI shown in Fig. 6 is the second layer screen of the current frame (see Fig. 2), and the screen GM2 is the second layer screen of the previous frame.
If it is a hierarchical screen, this motion vector detection unit 20
are the 8x8 pixel blocks BLI and BL2 of the screen GMI
A motion vector is sequentially detected for each of . Next, the process of detecting this motion vector will be explained. FIG. 7 (A) is a diagram showing each pixel of the screen GMI, and (B) is a diagram showing each pixel of the screen GM2. In this diagram, pixels DO and DOa, DI and D l m, . . .・-・、D
8 and DOa correspond to each other. First, the same process as the process of subtracting the image data of pixel DO from the image data of pixel Dla to find its absolute value is performed for all pixels in block BLI, and the results are sequentially accumulated. The minimum value detection unit 2 uses the result as first evaluation value data.
Output to 9. This first evaluation value data corresponds to the components of the vector Vl shown in FIG. Further, this process is performed when mx=l and my=0 in the equation (1).

Next, the same process as that of subtracting the image data of pixel Do from the image data of pixel D2a to find its absolute value is performed for all pixels in block BLI, and the results are sequentially accumulated. The calculation result is output to the minimum value detection section 29 as second evaluation value data. This second evaluation value data corresponds to the component of vector v2 shown in FIG.
This is the case where = l and my= -1.

Thereafter, the same process is repeated, and thereby the first to eighth evaluation value data corresponding to each component of the vectors v1 to v8 shown in FIG. 7 are output to the minimum value detection section 29. Finally, the same process as the process of subtracting the image data of pixel DO from the image data of pixel Dla to find its absolute value is performed for all pixels in block BLl, and the results are accumulated. The result is output to the minimum value detection section 29 as ninth evaluation value data. This ninth evaluation value data indicates the extent to which the image does not move on screen GM2 and screen GMI. For example, if this ninth evaluation value data is "0", each image in block BLl is
This is exactly the same as the image in block BL la of M2 (not moving). Further, this process is performed when mx=O and B=0 in equation (1).

Next, the process of calculating the above-mentioned first to ninth evaluation value data will be specifically explained. When the image data of the current frame, the previous frame, the first layer image data, and the second layer image data are respectively written to the hierarchical image memories 16 and 17, the control unit 30 clears the initial vector latch 23,
Next, a start signal is output to the search vector generating section 24 and the address generating section 21. Search vector generator 2
When the start signal to 4 is supplied, the search vector generation unit 24 first outputs first address correction data corresponding to mx=1 and my=O. Since the output of the initial motion vector latch 23 is "0" at this time, this address correction data is sent to the adder 22 via the adder 25.
is supplied to the second input terminal of.

On the other hand, when the start signal is supplied to the address generation section 21, the address generation section 21 thereafter starts the first block BLI.
Address data indicating the address of the memory 16 where the image data (second layer image data) of each pixel (FIG. 6) is stored is sequentially output. This address data is transferred to the address terminal AD of the memory 16 via the address bus B3.
At the same time, the address correction data is added in the adder 22 and sent to the memory 1 via the address bus B4.
7 address terminal AD. As a result, the image data of each pixel of the block BLI is sequentially output from the memory 16, and the image data of the pixel Do of FIG. 7 is output from the memory 17.
The pixel data of each pixel in the relationship with the pixel Dla is sequentially output, and the output pixel data is sent to the evaluation value calculation unit 28.
supplied to

The evaluation value calculation unit 28 subtracts the output data of the memory 17 from the output data of the memory 16 every time image data is supplied from the memories 16 and 17 to obtain its absolute value, and successively accumulates the results. , and 64 data (8x8 pixels)
When the accumulation is completed, the result is outputted to the minimum value detection section 29 as first evaluation value data. Minimum value detection section 29
receives this evaluation value data, stores it in its internal memory, and then outputs a signal indicating completion of evaluation value calculation to the control unit 30. The control unit 30 receives the signal indicating completion of the evaluation value calculation and outputs a vector change command to the search vector generation unit 24. The search vector generation unit 24 receives this vector change command and outputs address correction data corresponding to 111=1 and tay=−1 to the adder 25.

On the other hand, the address generation unit 21 thereafter sequentially outputs address data indicating the address of the memory 16 where the image data of each pixel of the first block BLI is stored. As a result, the image data of each pixel of the block BLI is sequentially output again from the memory 16, and the memory 17
From there, the pixel data of each pixel in the relationship of pixel D2a to pixel DO in FIG. The evaluation value calculation unit 28 calculates second evaluation value data based on each supplied image data, and outputs it to the minimum value detection unit 29. The minimum value detection section 29 receives this evaluation value data, stores it in an internal memory, and then outputs a pulse signal to the control section 30. Hereinafter, the same process is repeated, and as a result, the first to
The ninth evaluation value data is stored in the memory within the minimum value detection section 29. Here, the minimum value detection unit 29 detects the minimum value among the first to ninth evaluation value data.

The vector corresponding to this detected evaluation value data is the motion vector of block BLI. For example, if the second evaluation value data is the minimum, the motion vector of block BLI is detected as vector V2 (FIG. 2). Further, when the ninth evaluation value data is the minimum, the motion vector of the block BLI is detected as "0".

Next, the minimum value detection unit 29 outputs data indicating the detected motion vector to the control unit 30. The control unit 30 sends the data indicating this motion vector to the search vector generation unit 24.
Output to. The search vector generating section 24 receives data indicating this motion vector and again outputs address correction data corresponding to the data. As a result, adder 25
The same address correction data is output from the output latch 26.
supplied to Here, the control unit 30 outputs a read command to the output latch 26. As a result, the address correction data is read into the output latch 26 and supplied to the motion vector memory 27. Here, the control unit 30
The read command and address are output to 7. As a result, address correction data corresponding to the motion vector detected by the minimum value detection section 29 is written into the memory 27.

Next, the control section 30 sends the block B to the address generation section 21.
Outputs a pulse signal indicating completion of LI processing. Thereafter, the motion vector of the next block BL2 is detected by the same process as described above, address correction data indicating the detected motion vector is stored in the motion vector memory 27, and then the motion vector of the next block BL3. Similar processing is sequentially performed for BL4, . . . as well.

The above is the motion vector detection process based on the second layer image data. When this process is completed, a motion vector detection process based on the first layer image data is then performed. This process is exactly the same as the process described above, except that the data in the initial vector latch 23 is different from the case described above. That is, in the motion vector detection process based on the second layer image data described above, the initial vector latch 23
However, in processing based on the first layer image data, the data at address %17 in the motion vector memory 27 is doubled (that is, shifted in the upward direction by l bits). ) is written into latch 23.

The details will be explained below. Reference numeral GM3 in FIG. 8 is a screen based on the image data of the first layer. The number of pixels in this screen GMa is four times the number of pixels in the screen GMI in FIG. 6, and four pixels in the screen GM3 correspond to one pixel in the screen GM1.

Also, the block BLI in FIG. 6 is the block BM in FIG.
1, l3M2, I3MK, and BM(K+1), and similarly, the other blocks in FIG. 6 also correspond to the four blocks in FIG. 8, respectively. The motion vector detection process based on the first layer image data is sequentially processed in the order of blocks BMI, BM2, etc. in FIG. 8, as in the case described above, but in the block BLI in FIG. The corresponding blocks BM I , 8M2 . BMK, BM(
In the process K+1), address correction data indicating the motion vector of block BLI is read from the memory 27, this address correction data is doubled, and written to the initial vector latch 23.

The meaning is as follows. Now, if the motion vector of block BLl is vector v2 in FIG. 7, that is, mx = 1, my = -1, then the vector is twice the vector ■2, that is, rAK' = IX2 =
2. The vector my'=-1x2=-2 is block B
MI, 8M2. BM- becomes the reference vector for calculating the motion vector of BM(K+1). Therefore, in FIG. 9, if pixel DO is a pixel corresponding to the first hierarchical image data in hierarchical image memory 16, pixel Dd shown in the figure becomes the reference for motion vector detection. In the motion vector detection process based on the first layer image data,
Any one of the motion vectors Vat to Va9 shown in FIG. 9 will be detected.

And all the blocks BMl, 8M2, etc. in Fig. 8.
When the motion vector detection for ... is completed and the address correction data corresponding to the detected motion vector (in this case, the address correction data output from the adder 25) is stored in the motion vector memory 27, the next In memory 16.
17 storage areas Ear.

A motion vector detection process is performed based on the original image data in EbO. This process is exactly the same as the motion vector detection process based on the first layer image data described above, and the first layer address correction data stored in the initial motion vector latch 27 is doubled and the initial motion vector latch is 23
and each block (8/
Address correction data corresponding to each motion vector detected for 8 pixels) is written into the motion vector memory 27. This written address correction data is read out from the memory 27 in accordance with the timing of image transmission, and is outputted from the output terminal 31 as motion vector data MV.

The above is the motion vector detection circuit 1 according to one embodiment of the present invention.
1 details.

Note that in the above embodiment, the hierarchical image generation unit i4,1
5, 1st. Although the second layer image data was generated at a reduction ratio of 1/2, other reduction ratios may be used. If the reduction rate is set to 1/4, for example, the motion vector memory 2
The address correction data written in 7 may be multiplied by 4 and written in the initial vector latch 23. Furthermore, in the above embodiment, the average value of the four image data of the current frame was used as the first layer image data, and the average value of the four pieces of first layer image data was used as the second layer image data. Instead, each image data may be multiplied by a weighting coefficient (for example, 0.4, 0.3, 0.2, 0.1) determined by the relative positional relationship of a plurality of pixels, and then added. .

FIG. 10 is a block diagram showing an example of the configuration of the hierarchical image generation section 14.15 in this case. In this figure, 33 is a coefficient generation unit that generates the above-mentioned weighting coefficients, 34 is a product-sum operation unit, 35 is a control unit, and 36 is an address generation unit that generates an address for the hierarchical image memory 37.
The image data sequentially read from 7 is multiplied by the coefficients output from the coefficient generating section 33 in the sum-of-products calculation section 34, and then sequentially added, and the addition result is written into the memory 37.

In addition, in the above embodiment, the motion vectors are obtained in order from the highest layer, but instead of this,
Motion vector detection may be started from any hierarchy. In this case, when detecting the motion vector of a target block in a certain hierarchy, if that hierarchy is the highest, the value of the initial motion vector latch 23 is cleared,
A motion vector detection operation is executed, but if the block is not at the top level, it is checked whether the motion vector of the block in the upper layer has already been detected, and if the detection has been completed, the motion vector is read out from the motion vector memory 27 and initialized. The data is written to the motion vector latch 23 and a motion vector detection operation is executed. If the detection is not completed, the motion vector detection operation is executed at a higher layer. The motion vector detected at the lowest layer is output as the detection result.

Furthermore, in the embodiment described above, the motion vector obtained using coarse hierarchical image data of a higher resolution may be offset in processing based on fine hierarchical image data of a lower resolution. In this way, the initial motion vector can be set to "0" in the processing of lower hierarchical image data.

Next, an image data transmission apparatus to which the motion vector detection circuit 11 according to the above embodiment is applied will be explained with reference to FIG. FIG. 11 is a principle block diagram of an encoding device and a decoding device using a motion compensated interframe predictive encoding method. In this figure, 11 is the above-mentioned motion vector detection circuit;
2 is a prediction signal generator, 43 is a subtracter, 44 is an encoder,
45 is a decoding section, 46 is an adder, 47 is a delay section, 48 is a transmitting section, 49 is a receiving section, 50 is a decoding section, 51 is an adder, 52 is a delay section, and 53 is a predicted signal generating section. .

In the encoding device, the human image data SkC (the same data as the image data CDI of the current frame described above) is applied to the subtracter 43 and the motion vector detection circuit 11. The motion vector detection circuit 11 calculates a motion vector MV for each block consisting of a plurality of pixels from the previous frame image data Rk-1 (same data as the above-mentioned image data GD2) output from the delay section 47 and the input image signal Sk. is detected and output to the prediction signal generation section 42 and the transmission section 48. The prediction signal generation unit 42 outputs the image data of the pixels of the previous frame that have been moved by the motion vector MV to the subtracter 43 and the adder 46 as prediction data Pk. The subtracter 43 subtracts the prediction data Pk from the input image data Sk to obtain difference data Dk.
is calculated and output to the encoding section 44. The encoding unit 44 quantizes and encodes the difference data Dk to generate a code 1k.
is calculated and outputted to the transmitter 48 and the trace encoder 45, and the decoder 45 calculates the reproduced difference data Ek from the code 1k and performs addition 4.
46. The adder 46 adds the reproduced difference data Ek and the predicted data Pk to obtain reproduced image data Rk, and outputs it to the delay unit 47. The delay unit 47 reproduces the reproduced image data Rk.
is delayed by a time interval of one frame, and the reproduced image data Rk-1 (GD2) of the previous frame is output to the motion vector detection section 11 and the prediction signal generation section 42. The transmitter 48 multiplexes the motion vector MV with code 1, modulates it, and transmits it to the decoding device.

In the decoding device, the receiving unit 49 demodulates the motion vector MV into the code 1 multiplexed and transmitted from the coding device, obtains the motion vector M into the code 1, and calculates the decoding FRS 50 and the prediction, respectively. The signal is output to the signal generating section 53. The decoding unit 50 obtains reproduced difference data Ek from the code Ik and outputs it to the adder 51.

The adder 51 adds the reproduced difference data Ek and the prediction data Pk to obtain reproduced/regenerated image data Rk, and outputs it as reproduced image data of the decoding device and to the n extension section 52. The delay unit 52 delays the reproduced image data rtk by a time interval of one frame and reproduces the reproduced image data Rk of the previous frame.
-1 to the prediction signal generation section 53, and the prediction signal generation section 5
3 outputs the reproduced image data of the previous frame shifted by the motion vector MV to the adder 51 as predicted data Pk.

In this transmission system, the amount of information to be transmitted strongly depends on the accuracy of the prediction data Pk, and the amount of information transmitted is strongly dependent on the accuracy of the prediction data Pk.
The role of the motion vector detection circuit 11 is important as well as the performance of the decoding section 50, and it is necessary to detect motion vectors in a wide search range with a small number of counts. Therefore, the present invention is extremely suitable for use in the above transmission system.

"Effects of the Invention" As explained above, the present invention first detects an approximate motion vector by generating hierarchical images from image signals, and then sequentially detects motion vectors with finer precision using lower hierarchical images. In hierarchical images, upper layers have fewer pixels than lower layers, and therefore the number of blocks with the same number of pixels is also small. This is a method that can significantly reduce the number of value calculations, and has the effect of reducing the number of evaluation value calculations by about 1/2 compared to the conventional method, that is, the three-stage tree search method.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the principle of this invention, FIG. 2 is a diagram showing an example of the structure of a hierarchical image, FIG. 3 is a block diagram showing the structure of an embodiment of this invention, and FIG. 4 is a hierarchical image generation unit 14. 5 is a diagram showing a current frame, FIG. 6 is a diagram showing a screen based on image data of the second layer, and FIG. 7 is for explaining motion vector detection in the second layer. , FIG. 8 is a diagram showing a screen based on the image data of the 1'th layer, FIG. 9 is a diagram for explaining motion vector detection in the 1st layer, and FIG. 1O is a diagram showing hierarchical image generation in FIG. 3. FIG. 11 is a block diagram showing another example of the configuration of the unit 14, FIG. 11 is a block diagram showing an application example of the present invention, and FIG. 12 is a diagram for explaining motion vectors. 11... Motion vector detection circuit, 14. [5...
...Hierarchical image generation unit, 16.17...Hierarchical image memory, 21...Address generation unit, 23.
... Initial vector latch, 24 ... Search vector generation section, 28 ... Evaluation value calculation section, 29
...Minimum value detection section. Applicant: Graphics Communication Co., Ltd. Figure 1 Figure 2 (Help f1A) Figure 6 Figure 7 Figure 8 Gm 3 Figure 9 oooo ○ 0 Figure 10

Claims (2)

[Claims] (1) In a motion vector detection method for detecting a motion vector indicating a moving direction and a moving distance of an image between a first frame and a second frame, the image data of the first frame and the second frame are detected. image data in the first and second memories, respectively, and
Converting the image data of the frame and the image data of the second frame into first to Kth hierarchical image data with sequentially coarser resolutions and storing them in the first and second memories, and storing the image data in the first and second memories with the coarsest resolution. The K-th hierarchical image data is
, read from the second memory to obtain a motion vector, then read the (K-1)th hierarchical image data from the first and second memories, and perform the above-mentioned motion vector calculation based on the read data. A motion vector is determined based on the result, and the same process is repeated.Finally, the image data of the first and second frames are read from the first and second memories, and based on the read data, the motion vector is calculated. A motion vector detection method, characterized in that a motion vector is determined based on a motion vector calculated from first hierarchical image data, and the determined result is output. (2) A motion vector detection method according to claim 1, in which a motion vector determined by higher resolution hierarchical image data with coarse resolution can be canceled out in processing based on lower resolution hierarchical image data with fine resolution.