JP2537243B2 - Motion compensation motion vector coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] The present invention relates to a motion compensation motion vector coding method in which an input motion vector of a detected input screen is variable-length coded by an entropy coder and then transmitted. For the purpose of reducing the number, a memory that stores a predetermined number of input motion vectors,
A codebook that stores a plurality of prediction error vector patterns obtained in advance, a predictor that generates a prediction motion vector, and a prediction error generator that generates a set of prediction error vectors from the input motion vector and the prediction motion vector And a matching determination circuit that compares the prediction error vector set with the prediction error vector pattern to select a prediction error vector pattern having the smallest error and sends the index to the entropy coder, the predictor comprising: The index output from the matching determination circuit is configured to generate the prediction motion vector from the prediction error pattern decoded by the decoder and the previous prediction motion vector.

[Industrial applications]

The present invention relates to an image coding system, and more particularly to a motion compensation motion vector coding system in which a detected input motion vector of an input screen is variable-length coded by an entropy coder and then transmitted.

In recent image transmission such as video conferencing, various proposals have been made from the viewpoint of realizing a low bit rate by compressing the transmission amount as much as possible in order to effectively use the transmission line and reduce the terminal cost. However, the so-called block matching type motion compensation method, which detects the motion of the transmission screen and obtains the difference signal by compensating the prediction screen by the amount of the motion and further quantizing and transmitting the difference signal, is in the limelight. Is coming.

[Conventional technology]

FIG. 12 shows a well-known coding system with motion compensation, which was announced at the Institute of Electronics and Communication Engineers on April 5, 1984, for example.

In the figure, 11 is a quantizer, 12 is a frame memory, 13 is a variable delay device, and 14 is a motion detector. In this method, as shown in FIG. 13, input screen block x (I, J) Absolute value of pixel-by-pixel difference between the same position on the reference screen stored in the frame memory 12 and the surrounding blocks y (i, j) | x (I, J) -y (i, j) | Is taken by the motion detector 14 and accumulated for one block, and the accumulated value Block is predicted to be the closest to the screen block before moving, that is, the predicted block is delayed by the amount moved by the variable delay unit 13 and the difference with the next input screen block is obtained. The quantizer 11 quantizes and sends out to the transmission path. At the same time, the prediction screen is stored in the frame memory 12 (via an inverse quantizer (not shown)) and is used to determine the next prediction block. The amount of motion of the detected block (hereinafter, referred to as a motion vector value) is separated from (or together with) the coding circuit of the quantized output of the quantizer 11 by the entropy coder (coding circuit) shown in FIG. ) Send to receiver via 10.

The receiving side reproduces the original screen from the received signal according to the motion detection information.

[Problems to be solved by the invention]

In the above-described conventional motion compensation motion vector entropy coder as shown in FIG. 14, a set of a motion vector in the X-axis (vertical axis) direction and a motion vector in the Y-axis (horizontal axis) direction (V _(X , V _Y ) is appropriately variable-length coded and transmitted individually to the receiving side, so although individual motion vectors can be transmitted accurately, the smaller the size of the block when performing motion compensation becomes The problem is that the amount of information increases due to motion compensation.

Therefore, it is an object of the present invention to reduce the amount of information transmitted by a motion compensation vector in a motion compensation vector encoding method in which an input motion vector of a detected input screen is variable length coded by an entropy coder and then transmitted. To do.

[Means for solving problems]

FIG. 1 conceptually shows a motion compensation motion vector encoding system according to the present invention for achieving the above object. 1 is a memory for accumulating a predetermined plurality of input motion vectors, and 2 is a preset memory. A codebook that stores a plurality of obtained prediction error vector patterns, 3 is a predictor that generates a prediction motion vector, and 4 is a prediction error that generates a set of prediction error vectors from the input motion vector and the prediction motion vector vessel,
Reference numeral 5 compares the prediction error vector set with the prediction error vector pattern to select a prediction error vector pattern having a minimum error and sets its index to the entropy coder.
A matching determination circuit to be sent to 10, and the predictor 3 generates the prediction motion vector from the prediction error pattern obtained by decoding the index output from the matching determination circuit 5 by the decoder 6 and the previous prediction motion vector. Is.

[Action]

In the motion compensation motion vector encoding method of the present invention shown in FIG. 1, the motion of the input screen with respect to the prediction screen is stored in the memory 1 as the input motion vector. The memory 1 stores a predetermined plurality of input motion vectors. Further, the codebook 2 stores a plurality of prediction error vector patterns obtained in advance in the X-axis direction and the Y-axis direction of the input motion vector. In addition, the predictor 3
A current prediction motion vector obtained from the prediction error pattern obtained by decoding the index output from the matching determination circuit 5 by the decoder 6 and the previous prediction motion vector is generated.

The prediction error generator 4 generates a prediction error vector set from the plurality of input motion vectors stored in the memory 1 and the prediction motion vector stored in the predictor 3, and uses this prediction error vector set as a matching determination circuit. Send to 5. The matching determination circuit 5 compares the above-mentioned prediction error vector set with the prediction error pattern stored in the codebook 2 in advance, selects the prediction error vector pattern with the smallest error, that is, the distortion, and determines its index as entropy. Send to Coder 10. The entropy coder 10 encodes the index into a variable length code, sends it to the transmission path, and sends it to the receiving side.

As described above, the prediction error vector pattern in the codebook that is closest to the prediction error vector set in the X-axis direction vector and the Y-axis direction by a plurality of combinations of the input motion vectors, that is, the prediction error vector closest to the group of the input motion vectors. The pattern is transmitted in the form of an index having a small amount of information.

〔Example〕

Embodiments of the motion compensation motion vector encoding method according to the present invention will be described below.

FIG. 2 shows an embodiment in which the motion compensation motion vector coding system of the present invention shown in FIG. 1 is incorporated into a motion compensation coding system. In this embodiment, a motion detector is used.
The input motion vector detected in 14 is stored in the memory 1 by a predetermined number. In the predictor 3, the motion vector that has already been transmitted, for example, the previous motion vector value in the X-axis and Y-axis directions is obtained and stored from the prediction error pattern of the decoder 6 and the previous prediction value. . Therefore, the prediction error generator 4
Then, the vector values stored in the memory 1 and the predictor 3 are input to generate a set of prediction error vectors for each of the X-axis direction and the Y-axis direction, and the set is sent to the matching determination circuit 5. The matching determination circuit 5 compares the prediction error vector pattern stored in the codebook 2 with the prediction error vector set in the X-axis direction and the Y-axis direction, selects the prediction error vector pattern with the smallest error, and sets its index. Entropy coder
Transmit to the receiver via 10.

The index of the selected prediction error vector pattern is also input to the decoder 6, and a prediction error vector pattern is obtained by inquiring the index into the codebook 2, and the prediction error of a predetermined number in this pattern is obtained. A vector obtained by adding the vector and the predicted value of the predictor 3 is obtained as a motion vector and given to the variable delay unit 13. The predictor 3 stores the motion vector given to the variable delay unit 13 as the current prediction motion vector, and prepares for the next prediction error occurrence. The variable delay unit 13 shifts the previous screen in the frame memory 12 to generate a predicted screen.

Since various examples can be considered for the prediction error vector pattern stored in the codebook 2, those examples will be described below together with the input motion vector.

FIGS. 3 (a) and 3 (b) show prediction error vector patterns stored separately in the X-axis direction and the Y-axis direction in the codebook 2, and n _X vectors are provided for the X-axis direction. collect vector pattern of prediction errors were Deki by the m _X-number (each 1 to m _X is each of the index of the vector pattern) is prepared, the vector for the Y-axis direction n _Y
Pieces collected vector pattern m _Y (each 1 to m _Y is each of the index of the vector pattern) are pieces prepared. These vector patterns are obtained in advance and stored, but even if n _x = n _y m _x = m _y , n _x ≠ n _y , m _x ≠
It may be m _y . However, in the case of n _x = n _y , for X-axis direction and Y
It is not necessary to separate codebook 2 for the axial direction,
In that case, the pattern configuration is the same as that shown in FIG.

When the prediction error vector pattern of the codebook 2 is determined in this way, correspondingly, the number of input motion vectors stored in the memory 1 is n _{x in the} X-axis direction input V _x and in the Y-axis direction input V _y. Is automatically determined as n _y .

FIG. 5 shows an embodiment in which actual numbers are applied to the embodiment of the code book of FIG. 4, and in this embodiment,
With n _x = n _y = 3, three vectors in the X-axis direction and vectors in the Y-axis direction are stored in the memory 1 for matching.

FIG. 6 shows an input motion vector and an embodiment (n _x = n _y = 3) in which the input motion vector is stored in the memory 1. In this embodiment, three target input motion vectors (shown by diagonal lines) are separately stored in the memory 1 in the X-axis direction and the Y-axis direction. However, the input motion vector may be stored in the memory 1 as it is and rearranged by the next prediction error generator 4 as shown in FIG.

On the other hand, the predictor 3 stores the vector value in the X-axis direction and the vector value in the Y-axis direction (prediction value vector value) that are one left. That is, the value of the already transmitted motion vector is stored in the predictor 3 as a predictive motion vector to predict the input motion vector stored in the memory 1 and to obtain the prediction error.

Therefore, the prediction error generator 4 obtains the prediction error of the vector value (−6, 2) of the predictor 3 from the vector value from the memory 1 as shown in the figure and sends it to the matching determination circuit 5.

Next, an embodiment of the judgment operation in the matching judgment circuit 5 will be described with reference to the flow chart shown in FIG.

First, the prediction error vector from the prediction error generator 4 is V _I
(N) (n is the number of dimensions), the code of the prediction error vector pattern of the codebook 2 is V _O (n, m) (m is the number of codes and the index number), and variables for determination are D, R, and IDX.
(Index) (step S1 in FIG. 7).

As an initial value, R = ∞ (however, ∞ means the maximum value that the cumulative value D can take), J (count) = 0, D = 0, I
= 0 (step S2). Then, in step S3, J
Is incremented by "1".

In step S4, I is incremented by "1" and the absolute value error (represented by ABS) between V _I (I) and V _O (I, J) is calculated, and the value obtained by adding D to this is added. As D
Repeat steps S4 and S5 until J reaches n. As a result, in FIG. 4, the error between the prediction error vector values 1 to n of the prediction error vector pattern of the index “1” and the X-axis direction vector (or Y-axis direction vector) of the prediction error vector Is accumulated.

When the index "1" ends, R and D are compared (step S6 in the same step), and if D has not reached R, R
= D and set the current cumulative value D to the previous cumulative value D and ID
With X = J (step S7), steps S3 to S8 are repeated until J = m. As a result, the error is accumulated for each of the m prediction error vector patterns, and at the same time, the ID that gives a smaller D in steps S6 and S7.
In order to obtain X, the index of the prediction error vector pattern that gives the smallest error (distortion) is finally selected and sent to the entropy coder 10.

It should be noted that, in step S4 for obtaining the error, the square error (V _I (I) −V _O (I, J)) ² may be used instead of the absolute value.

8 (a) and 8 (b) are the same as FIG. 3 except that n _x = 3, _ny
FIG. 9 shows an example of the codebook when = 7, and FIG. 9 shows an example of the corresponding prediction error vector.

The reason why the number of prediction error vector patterns in the X-axis direction and the Y-axis direction is changed in this way is that the horizontal motion is usually larger than the vertical motion in the video conference. In this case, as shown in the figure, only three X-axis direction vectors of the input motion vector are processed, and waiting for input may occur (in order to prevent this, a buffer should be provided to wait until seven are accumulated). However, since seven Y-axis direction vectors are processed together, no input waiting occurs.

Unlike the above, FIG. 10 shows an embodiment in which the codes in the X-axis direction and the Y-axis direction are not separately stored but the codes in both directions are paired and stored together. In this embodiment, the prediction is performed. As for the error vector, both the Z-axis direction vector and the Y-axis direction vector are subjected to the arithmetic processing of FIG. 7 at the same time. FIG. 11 shows an embodiment in which actual numerical values are applied.

On the receiving side, the reverse operation is performed to reproduce the motion vector.

That is, the prediction error vector pattern (this is the same codebook as the transmission side is prepared on the reception side) is extracted by the transmitted index, and the number of dimensions of this prediction error vector pattern (prediction error vector pattern It is sufficient to perform motion-compensated reproduction using the same prediction error vector as the number n) of vector.

Therefore, only the index of the matched motion vector pattern needs to be sent from the transmitting side to the receiving side.

〔The invention's effect〕

As described above, according to the motion compensation motion vector coding method of the present invention, a plurality of prediction error vector patterns are prepared, and a prediction error between a predetermined number of input motion vectors and prediction motion vectors is prepared. Since the one having the smallest error with respect to the prediction error vector pattern is selected and only the index is transmitted, a predetermined number of input motion vectors can be represented by one pattern.
Although there is some distortion, the amount of information is small and extremely efficient coded transmission can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing the principle of a motion compensation motion vector coding system according to the present invention, and FIG. 2 is an embodiment of a motion compensation coding system incorporating the motion compensation motion vector coding system according to the present invention. FIG. 3 is a block diagram showing an example of a codebook used in the present invention, FIG. 6 is a diagram showing an example of processing of an input motion vector, and FIG. 7 is a matching judgment. FIG. 8 is a flow chart for explaining the arithmetic operation of the circuit, FIG. 8 is a diagram showing an embodiment of a codebook used in the present invention, FIG. 9 is a diagram showing an embodiment of processing of an input motion vector, and FIG. FIG. 11 is a diagram showing an embodiment of a codebook used in the present invention, FIG. 11 is a diagram showing an embodiment of processing in the case of the codebook of the embodiment of FIG. 10, and FIG. 12 is a conventionally known motion compensation code. Fig. 13 is a block diagram showing an example of the conversion method. Diagram for explaining the principle of the feeder compensation, FIG. 14 is a diagram, showing an entropy coder for coding the conventional motion vector. In FIG. 1, 1 ... First memory, 2 ... Codebook, 3 ... Second memory, 4 ... Prediction error generator, 5 ... Matching determination circuit, 6 ... Decoder, 10 …… Entropy coder. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (5)

(57) [Claims] 1. A motion compensating motion vector coding system for variable-length coding and transmitting a detected input motion vector of an input screen by an entropy coder (10), and a memory for accumulating a predetermined plurality of the input motion vectors. (1), a codebook (2) that stores a plurality of prediction error vector patterns obtained in advance, a predictor (3) that generates a prediction motion vector, and a prediction from the input motion vector and the prediction motion vector A prediction error generator (4) that generates a set of error vectors is compared with the prediction error vector set and the prediction error vector pattern to select a prediction error vector pattern having the smallest error, and its index is set to the entropy coder. (1
And a prediction error pattern in which the predictor (3) composites the index output from the matching judgment circuit (5) by the compositer (6). A motion compensation motion vector coding method characterized in that the motion vector predictor is generated from the previous motion vector predictor. 2. The code book (2) has a Y-axis direction and a Y-axis direction.
The prediction error vector pattern is stored separately in the axial direction, and the matching determination circuit (5) divides the input motion vector into a set of values in the X-axis direction and a set of values in the Y-axis direction. The motion compensation motion vector encoding method according to claim 1, which is used for comparison with a prediction error vector pattern. 3. A motion compensation motion vector coding method according to claim 2, wherein the number of respective values of the prediction error vector pattern divided into the X-axis direction and the Y-axis direction is different. 4. The code book (2) has a Y-axis direction and a Y-axis direction.
The prediction error vector patterns divided in the axial direction are stored as a pair, and the matching determination circuit (5) forms a pair of a value in the X-axis direction and a value in the Y-axis direction of the input motion vector as a pair. The motion compensation motion vector encoding method according to claim 2, which is used for comparison with a vector pattern. 5. The codebook (2) has a Y-axis direction and a Y-axis direction.
The prediction error vector pattern common to the axial direction is stored, and the matching determination circuit (5) divides the input motion vector into a set of values in the X-axis direction and a set of values in the Y-axis direction. The motion compensation motion vector encoding method according to claim 2, which is used for comparison with a prediction error vector pattern.