KR100293445B1 - Method for coding motion vector - Google Patents

Abstract

PURPOSE: A motion vector coding method is provided to increase compression efficiency by minimizing a coding amount by selecting and coding a motion vector having a small amount of bit as a proposed motion vector. CONSTITUTION: A motion vector coding method includes the steps of defining two adjacent motion vectors having the nearest relationship as proposed motion vector, selecting one motion vector having a small amount of bit when coded as the final proposed motion vector, obtaining difference between the selected motion vector and a real motion vector, transmitting the obtained difference value of motion vector and predictive mode information to a coder, estimating a minimum bit rate, and coding the difference value of motion vector of an X element and a Y element using the minimum bit rate.

Description

Motion vector coding method {METHOD FOR CODING MOTION VECTOR}

The present invention relates to Moving Picture Experts Group (MPEG) -4. In motion vector coding, a bit rate according to video compression is selected by selecting a value closest to a real motion vector among two candidate motion vectors and coding the difference value and the mode bits. The present invention relates to a motion vector coding method suitable for improving the compression ratio by reducing the P s.

In general, in order to efficiently compress a video sequence that changes with time, it is absolutely necessary to remove redundancy on the time axis as well as redundancy on the two-dimensional space of the image data.

Moving Picture Experts Group (MPEG) uses a discrete cosine transform (DCT) to remove redundancy in two-dimensional space, and uses a motion compensation method to remove redundancy on a time axis.

The DCT is a method of removing the correlation of data through two-dimensional spatial transformation. In the spatial transformation of a picture, each block divided by a block unit is spatially transformed using the DCT.

Spatially transformed data tends to converge in one direction, and only the collected data is quantized and transmitted.

Since pictures that are continuous on the time axis mainly move people or objects in the center of the screen, the motion compensation method uses this property to remove the redundancy on the time axis.

In other words, the unchanged part of the screen (or the part that changes very little even if it is moved) can minimize the amount of data to be transferred by taking the similar part from the previous picture and filling it.

Finding the most similar block among pictures is called motion estimation and motion vector is the displacement vector.

In other words, a motion vector refers to a two-dimensional vector representing an offset of a coordinate of a reference frame or a reference field from coordinates of a current picture or a field for motion compensation of an image.

In the MPEC-4, which is currently being standardized, the 8x8 block search performed during the inner pixel motion estimation is performed within a search window of ± 2 pixels around a 16x16 motion vector. Perform.

Thus, when a particular macroblock is selected in 8x8 mode, each motion vector for the four blocks in the macroblock is all within a certain bound.

FIG. 1 shows a bound that a motion vector for four blocks of a macroblock selected in the 8 × 8 mode according to the prior art can point to.

The bounds of these four motion vectors greatly contribute to improving the coding efficiency of the motion vectors.

Up to four motion vectors can be obtained per macroblock. If they are simply transmitted, a large amount of bits can be used to transmit only the difference from the motion vector of all the macroblocks by variable length coding (VLC).

That is, since the motion vectors between adjacent macroblocks are highly correlated, coding a difference value with a previous motion vector can reduce the bit amount rather than coding individual motion vectors for each macroblock.

In addition, in order to more efficiently code a motion vector, MPEG-4, which is currently being standardized, does not simply obtain a difference value from a motion vector of a previous macroblock and codes the motion vector. As shown in Fig. 3, three candidate motion vectors are defined, and a method of coding a difference value with a motion vector of the current macroblock by selecting an intermediate value among them is used.

The reason for this is that when the motion vector coding method in MPEG-2 is used, the difference between the motion vector of the previous macroblock and the motion vector of the current macroblock becomes large in a portion where the correlation between the macroblocks is poor, such as a boundary of an image. This is because it lowers the coding efficiency.

Accordingly, in MPEG-4, three candidate motion vectors are selected, and an intermediate value is selected as a final candidate motion vector to obtain a difference value, thereby improving correlation between the motion vectors.

2 illustrates three candidate motion vectors for predicting a motion vector in a macroblock according to the prior art.

As shown in FIG. 2, the components of the motion vectors corresponding to each of the four blocks are encoded by using the median value of three adjacent candidate motion vectors as a prediction value.

If this is expressed as a formula, it is as follows.

P _X = Median (MV1 _X , MV2 _X , MV3 _X )

P _Y = Median (MV1 _Y , MV2 _Y , MV3 _Y )

MVD _X = MV _X -P _X

MVD _Y = MV _Y -P _Y

Here, the MVD indicates a difference value of the motion vector.

As shown in FIG. 2, when the motion vector prediction code in the 8 × 8 mode is used, if the motion vector value used as the median predictor value exceeds the motion vector limit in the 8 × 8 mode, the block vector is excluded from block 2 except for block 1. Only corresponding MVD ₂ is available.

In this case, since only one of the three motion vector prediction candidates belongs to the selected macroblock in 8x8 mode, the absolute value of MVD ₂ obtained from the prediction value obtained as the median value may exceed the limit.

As described above, a case in which the limit value is exceeded, that is, the predicted value due to the median value is outside the limit range, is illustrated in FIG. 3.

As described above, in order to predict one motion vector, the conventional motion vector coding method improves the correlation between motion vectors by selecting a median value of three adjacent motion vectors as a final candidate motion vector and obtaining a difference value.

However, the motion vector coding method according to the prior art has the following problems.

In order to minimize the difference of the motion vectors compared to the conventional coding method of MPEG-2, it is advantageous to minimize the difference of the motion vectors by selecting the intermediate value among the three adjacent candidate motion vectors as the final candidate motion vector. However, since uniformly selecting the median does not determine the minimum difference, there is a problem that there is a lot of motion and the coding efficiency is deteriorated in a complex image.

The present invention has been made to solve the above problems, by defining three to two candidate motion vectors, and one of them having a small bit amount in the coding of the motion vector, finally selected and coded as a candidate motion vector, It is an object of the present invention to provide a motion vector coding method that minimizes the coding amount and increases compression efficiency.

1 is a view showing a bound that can be pointed to the four block motion vectors of a macroblock selected in the 8 × 8 mode according to the prior art

2a to 2d are diagrams showing motion vector prediction candidates in an 8x8 mode macroblock according to the prior art.

3 is a configuration diagram illustrating an example in which a limit value is exceeded in a median prediction of a conventional motion vector.

4 is a diagram showing a motion vector prediction candidate according to the present invention.

5 is a bitstream syntax of MVD code bits and mode bits in motion vector coding according to the motion vector coding method of the present invention.

6a to 6b are diagrams illustrating an example of mode bit reduction in terms of encoder and decoder according to the motion vector coding method of the present invention.

7 is a flowchart for explaining a motion vector coding method of the present invention.

In the motion vector coding method of the present invention for achieving the above object, in the motion vector coding of MPEG-4, two adjacent motion vectors having the highest correlation are defined as candidate motion vectors, and then the two candidate motions defined above are defined. Selecting the one having the least bit amount as the final candidate motion vector in coding among the vectors, calculating the difference value between the selected final candidate motion vector and the actual motion vector, and transmitting the obtained motion vector difference value and the prediction mode information to the decoder. And estimating a minimum bit rate for coding the motion vector difference value, and then coding a motion vector difference value for the X component and the Y component using the minimum bit rate.

Hereinafter, a motion vector coding method of the present invention will be described with reference to the accompanying drawings.

First, according to the motion vector coding method according to the present invention, one of the two candidate motion vectors having the lowest bit rate for coding the motion vector is selected as the final candidate motion vector.

Then, the motion vector difference value (difference value between the actual motion vector and the defined motion vector) and the prediction mode information are sent.

4 illustrates two candidate motion vectors according to the present invention.

As shown in Fig. 4, one of two candidate motion vectors around the current motion vector is selected.

Here, MV is the current motion vector and MV1 is the previous motion vector. And MV2 represents the motion vector after.

Meanwhile, FIG. 5 illustrates a bitstream syntax of MVD code bits and mode bits according to the present invention.

Coding of the motion vector is performed on the X and Y components, respectively.

The mode bits indicate which candidates are used among two candidate motion vectors when MVD is coded.

Here, the MVD minimum bit rate prior to the mode bit can be reduced by using the MVD minimum bit information.

An example of such a reduction in mode bits is shown in FIG.

6A to 6B illustrate an example of mode bit reduction according to the motion vector encoding of the present invention in terms of an encoder and a decoder.

First, FIG. 6A illustrates the encoder side. When MV1 = 0, MV2 = 3 and the motion vector MV = 5 of the current block, MV2 closest to the MV2 far from MV2 is the final candidate motion. Select as a vector.

Accordingly, the minimum rate motion vector difference value MVD min rate is two.

As mentioned above, the encoder codes the MVD min rate by minimum rate prediction.

The encoder then checks with minimal information whether the decoder can detect the actual motion vector.

6B is shown from the decoder side, it can be seen that there are only two candidates in the embodiment of the present invention.

However, when decoding the candidate 1 with the MVD min rate received from the encoder, it can be seen that the candidate 1 is not a real motion vector because it is different from the actual MVD min rate.

Therefore, in this case, the encoder does not need to send a mode bit to the decoder to find the actual motion vector among the two candidate motion vectors.

In addition, even if two candidate motion vectors have the same value, it is not necessary to send an additional mode bit.

If the amount of bits generated when coding two motion vector difference values is the same, priority is given to one side, so that no additional mode bits need to be sent.

As described above, only when the decoding can be performed without the additional mode bits, the additional mode bits are not generated to suppress an unnecessary increase in the amount of bits.

Meanwhile, the motion vector X component using the minimum rate prediction according to the present invention is as follows.

MR MVD coding (MVx, f code, * vlc code mag, * residual, * bitstream)

{

/ * Find the minimum rate predictor between two neighboring candidates. * /

R1x = RATE (MVx-MV1x),

R2X = RATE (MVx-MV2x),

Rx min = MIN (R1x, R2x)

Px min rate = MVix corresponding to Rx min.

(At the same rate, the MV1 has the highest priority for Px min rate.)

MVDx min rate = MVx Px min rate

/ * Coding MVDx min rate. * /

MVD encoding (MVDx min rate, f code, vlc code mag, residual, bitstream);

}

The above algorithm is a coding method of the X component using the minimum rate prediction, and the coding method of the Y component is also the same as the coding method of the X component.

The overall motion vector coding method of the present invention as described above is as follows.

max = MAX (MV1x, MV2x),

min = MIN (MV1x, MV2x)

{/ * Minimum rate prediction is applied. * /

Minimum rate prediction and the corresponding MVDx min rate coding * /

MR MVD coding (MVx, f code, & vlc code mag, & residual, bs);

/ * Determine MODEx. * /

/ * Step1: Evaluate two MV candidates. * /

Candidate1x = MVD decoding (f code, vlc code mag, residual, MV1x);

Candidate2x = MVD decoding (f code, vlc code mag, residual, MV2x);

/ * Step2: Check if "vlc code mag ts" and "residual ts" resulting from Candidateix (i = 1,2) encoding, have the same values as "vlc code mag" and "residual", respectively. * /

candidate num = 0;

MR MVD coding (Candidate1x, f code, & vlc code mag ts, & residual ts, ts);

if (vlc code mag ts = vlc code mag && residual ts = residual)

candidate num ++;

MR MVD coding (Candidate2x, f code, & vlc code mag ts, & residual ts, ts);

if (vlc code mag ts = vlc code mag && residual ts = residual)

candidate num ++;

/ * Step3: Determine the MODEx and code it. * /

if (candidate num == 1)

No bit allocated for MODEx;

else if (candidate num == 2)

1 bit for MODEx

}

The motion vector coding method of the present invention according to the algorithm described above will be described with reference to a flowchart.

7 is a flowchart for explaining a motion vector coding method of the present invention.

As shown in Fig. 7, two adjacent motion vectors having the highest correlation are defined as candidate motion vectors (S701).

Thereafter, one of the two candidate motion vectors defined above having the smallest bit amount in coding is selected as the final candidate motion vector (S702).

Next, the difference value between the final candidate motion vector and the actual motion vector is obtained (S703), and the obtained motion vector difference value and the prediction mode information are transmitted to the decoder (S704).

After estimating the minimum bit rate for coding the motion vector difference value (S705), if the motion vector difference values for the X component and the Y component are coded using the minimum bit rate (S706), the motion vector coding according to the present invention is performed. This is done.

As described above, in the motion vector coding method of the present invention, after selecting the final candidate motion vector from two candidate motion vectors, the difference value and the mode bit are coded, so that the coding amount is reduced, thereby reducing the bit rate in moving picture compression. It is effective to increase.

Claims (3)

In motion vector coding of MPEG-4, Defining two adjacent motion vectors having the highest correlation as candidate motion vectors, and then selecting one of the two candidate motion vectors having the smallest bit amount when coding as the final candidate motion vector; Obtaining a difference value between the selected final candidate motion vector and the actual motion vector, and transmitting the obtained motion vector difference value and the prediction mode information to the decoder; And after estimating a minimum bit rate for coding the motion vector difference value, coding a motion vector difference value for an X component and a Y component using the minimum bit rate. The method of claim 1, And in the step of selecting the final candidate motion vector, if the two candidate motion vectors have the same value, i.e., the same bit quantity, additional mode bits are not sent in the future. The method of claim 1, In the step of coding the motion vector difference values for the X component and the Y component, if the bit amounts generated when coding the two motion vector difference values are the same, the priority is given to one side, and the additional mode bits are not sent. Characterized by the motion vector coding method.

Images