CN101035282A - AVS-M video coding fast motion estimation method based on the in-block down-sampling - Google Patents

Abstract

A block down sampling within the AVS-M Video Coding Fast Motion estimation methods, involving video compression technology, use of the invention in a block-matching criteria SAD video compression algorithm, on the block 16 X 16, 16 X 8 block, block 8 X 16 X 8 block 8 four blocks of a pattern of arbitrary block by block pixel sampling points within the interval sampling of pixels calculated manner, that is, within the pixel block with points in the matrix, by the way interval of each and every trip out of pixels in the sampling points. SAD relative to the whole sampling method, the invention will be sampling of the search algorithm greatly reduce computational complexity, lower rate could reach 50%; PSNR slightly decreased, but its visual quality almost no change; Bit rate slightly increased. Overall, the coding efficiency is greatly improved.

Description

AVS-M video coding fast motion estimation method based on in-block down-sampling

Technical field

The present invention relates to the video compression technology field in the multimedia application, particularly a kind of AVS-M video coding fast motion estimation method based on in-block down-sampling.Described " AVS-M " full name is " a digital audio/video mobile multimedia national standard ".

Background technology

Since C.E.Shannon in 1948 proposed the source encoding theory, people had carried out extensive studies to the image compression encoding technology.A large amount of video compression technologies and method have appearred in the research and development through nearly 60 years, comprise predictive coding, vector coding, transition coding, entropy coding, object-based shape coding and Coding with Wavelets, fractal image etc.According to the information source model that video-frequency compression method adopted, these compress techniques can be divided into two big classes: content-based coding method and based on the coding method of waveform.

Adopted the block-based hybrid coding method that predictive coding and transition coding are combined based on the coding of waveform.In order to reduce the complexity of coding, make the video coding operation be easy to carry out, when adopting the hybrid coding method, the piece that at first piece image is divided into fixed size, for example piece 8 * 8 (is every 8 row, 8 pixels of every row), piece 16 * 16 (every 16 row, 16 pixels of every row) or the like, then piece is carried out compressed encoding and handles.

Predictive coding is data compression one of the most effective means.Data dependence in the video in the two field picture makes and can utilize the data message of its adjacent block to predict for certain data block, only keeps the poor of original pixel and prediction pixel, thereby reduces or eliminate spatial redundancy, and this is an infra-frame prediction; Temporal correlation in the video between the adjacent image frame makes data block can utilize its adjacent image frame (reference frame) to predict, only keeps the poor of original pixel and prediction pixel, eliminates or reduces spatial coherence, and this is an inter prediction.

Usually, when realizing inter prediction, divide for each piece, need in reference frame, search the reference block that mates most in the corresponding hunting zone, determine that the criterion of the piece of coupling mainly contains three kinds of MSE (MeanSquared Error), MAE (Mean Absolute Error) and SAD (Sum of AbsoluteDifference), it defines respectively suc as formula shown in (1), (2) and (3) formula.

$\mathrm{MSE}=\frac{1}{N^2}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}{(C_{i,j}-R_{\mathrm{MV}(i,j)})}^2---\left(1\right)$

$\mathrm{MAE}=\frac{1}{N^2}\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}|C_{i,j}-R_{\mathrm{MV}(i,j)}|---\left(2\right)$

$\mathrm{SAD}=\underset{i=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}|C_{i,j}-R_{\mathrm{MV}(i,j)}|---\left(3\right)$

Wherein, C represents the value of the picture element in the frame of video to be encoded; R represents the value of the picture element in the reference frame; M, N represent picture element line number and the columns in the block size respectively, and i, j represent the row value that the capable value of first summation number and second sue for peace number respectively; (i j) is illustrated in the reference frame with respect to (i, offset j) MV.

Comparatively speaking, the SAD criterion is the simplest, H.261, H.263, H.264 with in the international food compression standards such as MPEG-1, MPEG-2, MPEG-4 all recommend to use the SAD criterion.Though adopt the SAD criterion simple than other criterions, in matching process, amount of calculation still is very large, this mainly be because following some:

1, in the piece matching process, each macro block (size is 16 * 16) (for example: 16 * 16,16 * 8,8 * 16,8 * 8) search for need be divided into multiple sub-block mode;

2, every kind of piece is divided and will be searched in the given range of reference frame, and this scope causes the match search number of times of each piece to increase along with the increase of search point;

3, motion search will carry out in more than a reference frame usually.

Video coding with CIF form (size is 352 * 288) is an example, if (for example: 16 * 16,16 * 8,8 * 16,8 * 8) consider four kinds of piece partition modes, hunting zone in reference frame is ± 32 picture elements, reference frame is 2 frames, for a frame video, realization need be done subtraction, 3449952 absolute value operation and the operation of 3449951 sub-additions of (22 * 18) * (33 * 33) * 4 * 2=3449952 time based on the motion search of SAD criterion, for frame per second is the video sequence coding of 25 frame/seconds, and amount of calculation is quite huge.

In existing motion search fast algorithm, normally by adopting certain stopping criterion early, omitting as much as possible in search procedure can not be the search point of Optimum Matching and the method for piece partition mode.Even do like this because the mode that still adopts picture element to sample entirely in piece is carried out computing, so its amount of calculation is still quite huge.Therefore, how reduce amount of calculation the search procedure from other angle, and the algorithm that is proposed can combine with existing rapid movement searching algorithm well, further to reduce computational complexity, be the problem that the present invention does heavily to study.

Summary of the invention

The invention provides a kind of AVS-M video coding fast motion estimation method based on in-block down-sampling, its objective is and guaranteeing that the coding quality prerequisite decline harmonic motion suitable with optimal searching algorithm searched for and the computation complexity of piece coupling, reduce the software time of implementation of search blocks and optimal matching blocks, realize higher code efficiency.

For achieving the above object, the technical solution used in the present invention is: a kind of AVS-M video coding fast motion estimation method based on in-block down-sampling, its innovation is: adopting the SAD criterion to carry out in the video compression algorithm of piece coupling, M * N pixel block partition mode for defined any one arranged, sample mode by interval sampling picture element in the piece calculates, promptly in piece in the picture element arranged, by mode at interval the picture element in every row and the every row is sampled, and calculates by following formula (A) or formula (B):

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{\frac{M}{2}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{\frac{N}{2}-1}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+\mathrm{1,2}j}-R_{\mathrm{MV}(2i+\mathrm{1,2}j)}\left|\right)---\left(A\right)$

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{\frac{M}{2}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{\frac{N}{2}-1}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+\mathrm{1,2}j+1}-R_{\mathrm{MV}(2i+\mathrm{1,2}j+1)}\left|\right)---\left(B\right)$

In the formula, C represents the value of the picture element in the frame of video to be encoded; R represents the value of the picture element in the reference frame; M, N represent picture element line number and the columns in the block size respectively, and i, j represent the row value that the capable value of first summation number and second sue for peace number respectively; (i j) is illustrated in the reference frame with respect to (i, offset j) MV.

It is one of following four kinds that the partition mode of pixel block described in the such scheme can be defined as:

(1), M * N pixel block partition mode is piece 16 * 16, i.e. every 16 row, 16 pixels of every row.

(2), M * N pixel block partition mode is piece 16 * 8, i.e. every 16 row, 8 pixels of every row.

(3), M * N pixel block partition mode is piece 8 * 16, i.e. every 8 row, 16 pixels of every row.

(4), M * N pixel block partition mode is piece 8 * 8, i.e. every 8 row, 8 pixels of every row.

Main thought of the present invention is to adopt the SAD criterion to carry out in the video compression algorithm of piece coupling, by the down-sampled computational complexity that reduces of pixel in the piece.In the motion estimation algorithm of existing employing SAD criterion, all picture elements during each piece is divided all participate in computing, and in the present invention, calculate by the interval sampling picture element.

Because the technique scheme utilization, the present invention compared with prior art has following advantage and effect:

1, the present invention is to effectively reduce amount of calculation under the prerequisite of the framework that does not change the video coding flow process with respect to full sampling SAD method, has therefore still kept the characteristics of SAD.

2, the present invention is with respect to full sampling SAD method, reduces greatly based on the computational complexity of down-sampled searching algorithm, and the reduction amplitude can reach 50%; Y-PSNR PSNR decreases slightly, but its visual quality does not almost change; Code check increases slightly to some extent.Generally, code efficiency improves a lot.

3, this method is applied to block type in AVS-M (the digital audio/video mobile multimedia national standard) standard code be 16 * 16,16 * 8,8 * 16,8 * 8 piece divide motion search and with the full sampling SAD method contrast test that experimentizes, see Fig. 5～Fig. 8, can verify advantage of the present invention and good effect.

Description of drawings

Accompanying drawing 1 (a) is first kind of down-sampled schematic diagram in picture element interval of piece 16 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 1 (b) is second kind of down-sampled schematic diagram in picture element interval of piece 16 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 2 (a) is first kind of down-sampled schematic diagram in picture element interval of piece 16 * 8, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 2 (b) is second kind of down-sampled schematic diagram in picture element interval of piece 16 * 8, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 3 (a) is first kind of down-sampled schematic diagram in picture element interval of piece 8 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 3 (b) is second kind of down-sampled schematic diagram in picture element interval of piece 8 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 4 (a) is first kind of down-sampled schematic diagram in picture element interval of piece 8 * 8, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 4 (b) is second kind of down-sampled schematic diagram in picture element interval of piece 8 * 8, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.

Accompanying drawing 5～Fig. 8 is the contrast test of the present invention and full sampling SAD algorithm figure as a result.Wherein, coding standard is AVS mobile video standard A VS-M, and coded sequence selects four representative typical cases to look type sequence (generally using) in test: News, Foreman, Tempete and Football.The coding frame number is 250 frames, and frame per second was 30 frame/seconds.Coding efficiency curve when the coding efficiency curve when full SAD method is adopted in " Original " expression among the figure, " Rapid " expression have adopted down-sampled SAD method coding.

Embodiment

Below in conjunction with drawings and Examples the present invention is further described:

Embodiment one:

Fig. 1 (a) is first kind of down-sampled schematic diagram in picture element interval of piece 16 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.Shown in Fig. 1 (a), carry out in the video compression algorithm of piece coupling in employing SAD criterion, for defined 16 * 16 (is every 16 row, 16 pixels of every row) pixel block partition mode, sample mode by interval sampling picture element in the piece calculates, promptly in piece, in the picture element arranged, by mode at interval the picture element in every row and the every row is sampled, and calculates by following formula (A):

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{\frac{M}{2}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{\frac{N}{2}-1}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+\mathrm{1,2}j}-R_{\mathrm{MV}(2i+\mathrm{1,2}j)}\left|\right)---\left(A\right)$

In the formula, C represents the value of the picture element in the frame of video to be encoded; R represents the value of the picture element in the reference frame; M, N represent picture element line number and the columns in the block size respectively, and i, j represent the row value that the capable value of first summation number and second sue for peace number respectively; (i j) is illustrated in the reference frame with respect to (i, offset j) MV.

Because upper limit M, the N of first and second summations number be value 16 respectively, formula (A) becomes:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+\mathrm{1,2}j}-R_{\mathrm{MV}(2i+\mathrm{1,2}j)}\left|\right)---(A-1)$

From above formula (A-1) as can be seen, Fig. 1 (a) sampling picture element has only been selected half of piece 16 * 16 whole picture elements, and the scope of choosing is: row makes up so that parity is different with the row coordinate, at once with the row coordinate in, one is odd number, another is an even number, and its core is both parity differences.

Fig. 1 (b) is second kind of down-sampled schematic diagram in picture element interval of piece 16 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.Shown in Fig. 1 (b), carry out in the video compression algorithm of piece coupling in employing SAD criterion, for defined 16 * 16 pixel block partition mode, in picture element arranged in piece, by mode at interval the picture element in every row and the every row is sampled, and calculates by following formula (B):

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{\frac{M}{2}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{\frac{N}{2}-1}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+\mathrm{1,2}j+1}-R_{\mathrm{MV}(2i+\mathrm{1,2}j+1)}\left|\right)---\left(B\right)$

In the formula, C represents the value of the picture element in the frame of video to be encoded; R represents the value of the picture element in the reference frame; M, N represent picture element line number and the columns in the block size respectively, and i, j represent the row value that the capable value of first summation number and second sue for peace number respectively; (i j) is illustrated in the reference frame with respect to (i, offset j) MV.

Because upper limit M, the N of first and second summations number be value 16 respectively, formula (B) becomes:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+\mathrm{1,2}j+1}-R_{\mathrm{MV}(2i+\mathrm{1,2}j+1)}\left|\right)---(B-1)$

From above formula (B-1) as can be seen, Fig. 1 (b) sampling picture element has only been selected half of piece 16 * 16 whole picture elements, and the scope of choosing is: row makes up so that parity is identical with the row coordinate, at once with the row coordinate in, both are odd number simultaneously, both are even number simultaneously, and its core is that both parities are identical.

Fig. 1 (a) with the difference of Fig. 1 (b) is: in the picture element arranged, the sampling picture element of corresponding row and respective column is opposite in piece.

Embodiment two:

With the difference of embodiment one be: defined pixel block partition mode is a piece 16 * 8, i.e. every 16 row, 8 pixels of every row.

Fig. 2 (a) and Fig. 2 (b) are respectively first kind and second kind of down-sampled schematic diagram in picture element interval of piece 16 * 8, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.Fig. 2 (a) with the difference of Fig. 2 (b) is: in the picture element arranged, the sampling picture element of corresponding row and respective column is opposite in piece.

The formula that Fig. 2 (a) adopts is:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{3}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+\mathrm{1,2}j}-R_{\mathrm{MV}(2i+\mathrm{1,2}j)}\left|\right)---(A-2)$

The formula that Fig. 2 (b) adopts is:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{7}{\mathrm{\Σ}}}\underset{j=0}{\overset{3}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+1,2j+1}-R_{\mathrm{MV}(2i+\mathrm{1,2}j+1)}\left|\right)---(B-2)$

Other is identical with embodiment one, no longer is repeated in this description here.

Embodiment three:

With the difference of embodiment one be: defined pixel block partition mode is a piece 8 * 16, i.e. every 8 row, 16 pixels of every row.

Fig. 3 (a) and Fig. 3 (b) are respectively first kind and second kind of down-sampled schematic diagram in picture element interval of piece 8 * 16, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.Fig. 3 (a) with the difference of Fig. 3 (b) is: in the picture element arranged, the sampling picture element of corresponding row and respective column is opposite in piece.

The formula that Fig. 3 (a) adopts is:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{3}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+\mathrm{1,2}j}-R_{\mathrm{MV}(2i+1,2j)}\left|\right)---(A-3)$

The formula that Fig. 3 (b) adopts is:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{3}{\mathrm{\Σ}}}\underset{j=0}{\overset{7}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+\mathrm{1,2}j+1}-R_{\mathrm{MV}(2i+\mathrm{1,2}j+1)}\left|\right)---(B-3)$

Other is identical with embodiment one, no longer is repeated in this description here.

Embodiment four:

With the difference of embodiment one be: defined pixel block partition mode is a piece 8 * 8, i.e. every 8 row, 8 pixels of every row.

Fig. 4 (a) and Fig. 4 (b) are respectively first kind and second kind of down-sampled schematic diagram in picture element interval of piece 8 * 8, and the point of black represents to participate in the picture element of SAD computing among the figure, and colourless point does not participate in computing.Fig. 4 (a) with the difference of Fig. 4 (b) is: in the picture element arranged, the sampling picture element of corresponding row and respective column is opposite in piece.

The formula that Fig. 4 (a) adopts is:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{3}{\mathrm{\Σ}}}\underset{j=0}{\overset{3}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+1,2j}-R_{\mathrm{MV}(2i+\mathrm{1,2}j)}\left|\right)---(A-4)$

The formula that Fig. 4 (b) adopts is:

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{3}{\mathrm{\Σ}}}\underset{j=0}{\overset{3}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+\mathrm{1,2}j+1}-R_{\mathrm{MV}(2i+\mathrm{1,2}j+1)}\left|\right)---(B-4)$

Other is identical with embodiment one, no longer is repeated in this description here.

The foregoing description only is explanation technical conceive of the present invention and characteristics, and its purpose is to allow the personage who is familiar with this technology can understand content of the present invention and enforcement according to this, can not limit protection scope of the present invention with this.All equivalences that spirit is done according to the present invention change or modify, and all should be encompassed within protection scope of the present invention.

Claims (5)

1, a kind of AVS-M video coding fast motion estimation method based on in-block down-sampling, it is characterized in that: adopting the SAD criterion to carry out in the video compression algorithm of piece coupling, M * N pixel block partition mode for defined any one arranged, sample mode by interval sampling picture element in the piece calculates, promptly in piece in the picture element arranged, by mode at interval the picture element in every row and the every row is sampled, and calculates by following formula (A) or formula (B):

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{\frac{M}{2}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{\frac{N}{2}-1}{\mathrm{\Σ}}}\left(\right|C_{2i,2j+1}-R_{\mathrm{MV}(2i,2j+1)}|+|C_{2i+\mathrm{1,2}j}-R_{\mathrm{MV}(2i+\mathrm{1,2}j)}\left|\right)---\left(A\right)$

$\mathrm{SA}{D}^{\′}=\underset{i=0}{\overset{\frac{M}{2}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{\frac{N}{2}-1}{\mathrm{\Σ}}}\left(\right|C_{2i,2j}-R_{\mathrm{MV}(2i,2j)}|+|C_{2i+\mathrm{1,2}j+1}-R_{\mathrm{MV}(2i+1+\mathrm{1,2}j+1)}\left|\right)---\left(B\right)$

In the formula, C represents the value of the picture element in the frame of video to be encoded; R represents the value of the picture element in the reference frame; M, N represent picture element line number and the columns in the block size respectively, and i, j represent the row value that the capable value of first summation number and second sue for peace number respectively; (i j) is illustrated in the reference frame with respect to (i, offset j) MV.

2, rapid motion estimating method according to claim 1 is characterized in that: described M * N pixel block partition mode is a piece 16 * 16, i.e. every 16 row, 16 pixels of every row.

3, rapid motion estimating method according to claim 1 is characterized in that: described M * N pixel block partition mode is a piece 16 * 8, i.e. every 16 row, 8 pixels of every row.

4, rapid motion estimating method according to claim 1 is characterized in that: described M * N pixel block partition mode is a piece 8 * 16, i.e. every 8 row, 16 pixels of every row.

5, rapid motion estimating method according to claim 1 is characterized in that: described M * N pixel block partition mode is a piece 8 * 8, i.e. every 8 row, 8 pixels of every row.

Images