CN101668207A - Video coding switching system from MPEG to AVS - Google Patents

Abstract

The invention discloses a video coding switching system from MPEG to AVS. A coding transcoder of the system completely decodes MPEG video data for demultiplexed output to a pixel region and performs AVS coding by combining a DCT intra-frame predicative mode judgment module and a motion vector reuse and intra-frame macro block mode judgment module and is then transmitted to an MPEG system layer multiplexing module; the DCT intra-frame predicative mode judgment module in a converter decodes by utilizing VLC of the input code stream and performs intra-frame predicative mode judgment with the DCTcoefficient after inverse quantization; the motion vector reuse submodule in the motion vector reuse and intra-frame macro block mode judgment module uses the motion vector MV in the MPEG code streamas the reference of AVS motion estimation to obtain half-pixel precision motion vector MV after optimization of the motion vector MV; and the intra-frame macro block mode judgment submodule decodes the macro block through the MPEG and judges the coding type of the AVS macro block to divide the size. The invention avoids complex decoding-recoding processing, requires extremely low calculated amount, can realize MPEG-AVS transcoding in a software manner, provides AVS program source and promotes the application development of AVS standard.

Description

MPEG is to the AVS video coding conversion system

Technical field

The present invention relates to a kind of video coding technique, particularly relate to the system that utilizes software approach to realize the conversion of two kinds of video codings.

Background technology

The digital multimedia technical development is rapid, and the multimedia application field presents the situation of multiple standards coexistence.AVS (Audio Video coding Standard) be by science and technology department of China's Ministry of Information Industry in the standard that the audio/video encoding/decoding technical standard working group that in June, 2002 approval is set up is formulated, have the autonomous intellectual property of China.The AVS standard will obtain to use widely in fields such as Digital Television, IPTV and mobile multimedia service.

Yet because the industrialization of AVS standard also is in the starting stage, the digital multimedia resource that meets the AVS standard is less, is badly in need of providing the hardware or the software product of AVS form in the practical application.

Summary of the invention

In view of above-mentioned technical problem, the present invention proposes a kind of MPEG to the AVS video coding conversion system, utilize software to realize the conversion of mpeg encoded to the AVS coding, realize that expeditiously the code stream of a large amount of mpeg formats is shared as the AVS form.

The present invention proposes the video coding conversion system of a kind of MPEG to AVS, the data flow, mpeg system layer demultiplexing module and the mpeg system layer Multiplexing module that comprise mpeg format, it is characterized in that, this system also comprises coded conversion device, and the DCT intra prediction mode judge module in this coded conversion device, motion vector is reused and the inter macroblock judge module, the data flow of mpeg format is at first passed through mpeg system layer demultiplexing module, extracts MPEG video, voice data and other data behind the demultiplexing respectively; Wherein the mpeg audio data that obtain of demultiplexing are not carried out transcoding and are handled, directly send into " the mpeg system layer is multiplexing " module of rear end, and the MPEG video data that the coding transcoder is exported demultiplexing is fully decoded to pixel domain, again in conjunction with DCT intra prediction mode judge module, motion vector is reused and the inter macroblock judge module carries out AVS coding, send into mpeg system layer Multiplexing module then, the DCT intra prediction mode judge module in the transducer utilizes decode again DCT coefficient behind the inverse quantization of the VLC of input code flow to carry out intra prediction mode and judge; Motion vector reuse and the inter macroblock judge module in motion vector reuse submodule with the reference of the motion vector MV in the MPEG code stream as the AVS estimation, after motion vector MV optimization, obtain half-pixel accuracy motion vector MV, inter macroblock judges that submodule is then by the mpeg decode macro block, judge AVS macroblock encoding type, divide size.

The algorithm of described DCT intra prediction mode judge module may further comprise the steps:

In intra prediction mode is judged, angle θ of computed image texture (please replenish computing formula) and parameter ρ, can obtain current block to be encoded with and a left side is adjacent and the grain direction of last adjacent data block, when the adjacent or last adjacent data macro grain direction in a macroblock to be encoded and a left side is consistent, obtain intra prediction mode;

$\mathrm{\θ}=\arctan \left(\frac{\underset{u=1}{\overset{7}{\mathrm{\Σ}}}\left|F(u,0)\right|}{\underset{v=1}{\overset{7}{\mathrm{\Σ}}}\left|F(0,v)\right|}\right)---\left(1\right)$

In the formula, F (u, 0) is the first row AC coefficient of DCT coefficient; F (0, v) be the first row AC coefficient of DCT coefficient.When the adjacent or upward adjacent data macro grain direction in a macroblock to be encoded and a left side exists than big-difference, be that macroblock to be encoded, a left side are adjacent and go up and have two to satisfy 20 °＜=θ＜=35 ° or 55 °＜=θ＜=70 ° respectively and can not carry out level, vertical and oblique prediction the time in the adjacent macro block, need further to calculate prediction matrix in the DCT territory, the DCT data absolute error and the SAD that compare the current block to be encoded of intra prediction mode and current data block, obtain the optimum frame inner estimation mode, concrete algorithm process flow process is as follows:

Step 1 judges whether current block to be encoded is in image left hand edge or top edge, if then be judged as the DCT intra prediction mode, otherwise enter step 2;

Step 2 is judged current macro, and whether the θ value of left adjacent piece and last adjacent piece all satisfies T1＜θ＜T2, and T1=35 °, T2=55 ° are if illustrate that the current macro region has certain oblique texture, carry out step 3, otherwise carry out step 4;

Step 3 is judged the value of parameter ρ, if ρ=1 then can judge that the intra prediction mode of current block to be encoded is upper right intra prediction mode, ρ=0 is judged as the bottom right intra prediction mode;

Step 4 judges whether the θ value of current macro and last adjacent piece all satisfies 0 °＜=θ＜T3 (T3=20 °), if illustrate that the current macro region has certain vertical texture, can be judged as the vertical frame inner estimation mode, otherwise carry out step 5;

Whether step 5, the θ value of judging current macro and left adjacent piece all satisfy T4＜θ＜=90 °, and T4=70 °, if illustrate that the current macro region has the certain level texture, can be judged as horizontal intra prediction mode, otherwise carry out step 6;

Step 6, calculate respectively itself and current block to be encoded DCT data absolute error and:

$\mathrm{SAD}=|\mathrm{DCT}\left(D\right)-\mathrm{DCT}\left(S\right)|=\underset{i=0}{\overset{i<8}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{j<8}{\mathrm{\&Sigma;}}}|D_{\mathrm{ij}}-S_{\mathrm{ij}}|$

Step 7, note SAD _HZFor the absolute error of horizontal forecast and, SAD _VTFor the absolute error of vertical prediction and, SAD _DCFor the absolute error of DC prediction and.By comparing SAD _HZ, SAD _VTAnd SAD _DC, determine that the pairing pattern of reckling is the optimum frame inner estimation mode;

Above-mentioned parameter parameter ρ determines that by the sign symbol of the data of [0] [1] of 8 * 8 DCT coefficient matrixes and position, [1] [0] when two coefficient jack per lines, parameter ρ=1 is when two coefficient contrary signs, parameter ρ=0.

Described motion vector is reused and the inter macroblock judge module, this module mainly adopts the macro block level of detail as judgment criterion when judging macro-block partition mode, the inter macroblocks partition mode is selected 16 * 16 and 8 * 8 two kinds, and concrete deterministic process may further comprise the steps:

Step 1, all inter macroblocks all are initialized as 16 * 16 macro-block partition modes;

Step 2, the VLC number of coded bits of each the macro block residual error coefficient in the statistics MPEG code stream, calculate the VLC coding average number of bits of present frame macro block residual error coefficient, and whether belong to the basic thresholding of details as macro block than the judgment criteria of multizone with this average number of bits;

Step 3 is divided by with the VLC number of coded bits and the basic thresholding of each macro block residual error coefficient, if ratio greater than 1.5 times (experiment setting thresholds), the level of detail that this macro block then is described needs to use 8 * 8 patterns to encode obviously greater than the image averaging value; Otherwise use 16 * 16 patterns of acquiescence.

Motion vector reuse and the inter macroblock judge module in inter macroblock judge the algorithm of submodule, further comprising the steps of:

Judge AVS macroblock encoding type according to region characteristic, table 1, table 2 be in the transform coding process between mpeg frame macroblock encoding type to the conversion initialization mapping table of AVS inter macroblocks type of coding;

Table 1 P frame macroblock encoding type initialization mapping table

Table 2 B frame macroblock encoding type initialization mapping table

48 * 8 of B frame macro block pieces adopt different prediction direction to encode among the AVS, in order to obtain the type of coding of 8 * 8 sub-pieces, need further handle:

The type of coding of initialization 8 * 8 sub-piece E is the mpeg encoded type of the sub-piece of E place macro block, and current coding macro block is a forward predictive coded among the MPEG, and then the sub-initialization block of E is SB_Fwd_8 * 8; Obtain the mpeg encoded type of the sub-piece adjacent macroblocks of E A, B, C: if the initialization codes type of the sub-piece of E is SB_Fwd_8 * 8 or SB_Sym_8 * 8, and the MPEG macroblock encoding type of A, B, C three macro blocks is the back forecast coding, the type of coding of then judging the sub-piece of E is SB_Bck_8 * 8, otherwise is judged as the initialization codes type of E; If the initialization codes type of the sub-piece of E is SB_Bck_8 * 8 or SB_Sym_8 * 8, and the MPEG macroblock encoding type of A, B, C three macro blocks is forward predictive coded, the type of coding of then judging the sub-piece of E is SB_Fwd_8 * 8, otherwise is judged as the initialization codes type of E;

The step of inter macroblock evaluation algorithm is as follows:

Step 1, if the mpeg decode macro block corresponding with current coding macro block is intraframe predictive coding, then this macro block adopts the AVS intraframe predictive coding, because around the current macro all is inter macroblocks, has only residual error data, so can not use DCT territory intra prediction mode to judge as prosthomere, directly select the DC pattern here;

Step 2, the division size of judgement inter macroblocks: if 16 * 16 division sizes then carry out steps 3; If 8 * 8 division sizes then carry out steps 4;

Step 3, the type of coding that carries out under 16 * 16 partition modes according to table 1, table 2 shines upon;

Step 4, carry out the type of coding initialization mapping of 8 * 8 partition modes according to table 1, table 2, obtain the final type of coding of all 8 * 8 sub-pieces in the current coding macro block then according to 8 * 8 subblock coding type bearing calibrations of introducing previously, and encode expense relatively with 16 * 16 partition modes, select reckling to encode;

Step 5 is proofreaied and correct if B frame macro block then carries out the Direct pattern;

Step 6 is carried out intraframe predictive coding to the inter macroblocks that satisfies infra-frame prediction macroblock coding correcting condition and is proofreaied and correct, and the intra prediction mode decision method is with step 1.

Motion vector reuse and the inter macroblock judge module in motion vector reuse the algorithm of submodule, further comprising the steps of:

For adopting 16 * 16 coding mode and adopting 8 * 8 macro-block partition mode, all with the half-pixel accuracy motion vector of mpeg decode motion vector as optimum, the motion vector of doing one time 1/4 pixel precision on this motion vector basis is heavily searched for optimization then, promptly search for this half-pix motion vector 8 1/4 pixel precision motion vectors on every side, thereby obtain the motion vector of optimum 1/4 pixel precision, and 4 motion vectors of macro-block partition mode record that adopt 8 * 8.

Compared with prior art, the invention provides the transform coding system of a kind of MPEG, avoided complex decoding-recoding processing to AVS, required amount of calculation is extremely low, can adopt software mode to realize the MPEG-AVS transcoding, AVS is provided program source, advance the application development of AVS standard.

Description of drawings

Fig. 1 is the video coding conversion system structure chart of MPEG of the present invention to AVS;

Fig. 2 is an image texture deflection schematic diagram of the present invention;

Fig. 3 is 8 * 8 sub-piece time domain prediction direction schematic diagrames;

Fig. 4 is that 1/4 pixel motion vector is optimized schematic diagram;

Fig. 5 is the standard testing code stream experiment R_D curve chart of the embodiment of the invention;

Fig. 6 is the DVD sequence R_D curve chart of the embodiment of the invention.

Embodiment

At present, the digital multimedia resource in the practical application is mainly the mpeg standard form.The multimedia resource of mpeg standard can be divided into two kinds of program stream (PS) and transport stream (TS), at these two kinds of forms, can adopt transform coding structure shown in Figure 1 to finish transcoding and handle.But because the system layer and the audio standard of the equal MPEG compatible standard of system layer part in the AVS standard and audio coding standard, in the code conversion code system proposed by the invention to audio frequency and the multiplexing technology that meets mpeg standard that adopted of system layer.

As shown in Figure 1, be the video coding conversion system structure chart of MPEG to AVS, the digital multimedia program of mpeg format at first passes through " mpeg system layer demultiplexing " module 11, extracts MPEG video, voice data and other data behind the demultiplexing respectively; Wherein the mpeg audio data that obtain of demultiplexing are not carried out transcoding and are handled, and directly send into " the mpeg system layer is multiplexing " module 13 of rear end, and the MPEG video data of demultiplexing output are the AVS form by transcoding, sends into " the mpeg system layer is multiplexing " module 13 then.The mpeg format video data is seen frame of broken lines part among Fig. 1 to the structure of the coding transcoder 12 of AVS format video data, comprise that DCT intra prediction mode judge module 121 and motion vector MV reuse and frequency-domain macro block mode judge module 122, also comprise hardware components, comprise variable length decoding (VLD), inverse quantization (IQ), inverse discrete cosine transducer (IDCT), motion compensation (MC) etc.This transcoder 12 is fully decoded to pixel domain with the MPEG code stream of input, carry out the AVS coding again, crucial processing module in the transducer is a DCT intra prediction mode judge module 121, and DCT intra prediction mode judge module 121 utilizes decode again DCT coefficient behind the inverse quantization of the VLC of input code flow to carry out intra prediction mode and judge; Motion vector reuse and inter macroblock judge module 122 in motion vector reuse submodule with the reference of the motion vector MV in the MPEG code stream as the AVS estimation, after motion vector MV optimization, can obtain half-pixel accuracy MV comparatively accurately, macro block mode judgement submodule is then judged AVS macroblock encoding type fast by information such as mpeg decode macroblock encoding type, number of coded bits, divide parameters such as size, adopt this two parts technology can greatly reduce the transcoding amount of calculation.Different with traditional simple cascade transcoding structure (decoder and encoder cascade) is, code conversion shown in Figure 1 has made full use of mpeg decode information, before encoding certain macro block, AVS just determined this macroblock encoding parameter substantially, direct coding in the AVS encoder, rather than the coding of traversal formula, so, improved the efficient of code conversion.

Below will describe in detail to the main algorithm module among the present invention-" judgement of DCT intra prediction mode " and " motion vector is reused and inter macroblock is judged ".

One, DCT territory intra prediction mode is judged

The infra-frame prediction luminance frame inner estimation mode of AVS standard has 5 kinds, is respectively level, vertical, DC intra prediction mode, upper right and bottom right intra prediction mode.In fact, different intra prediction mode is applicable to the video data block of different texture characteristic.This patent adopts following algorithm to judge AVS intra prediction mode in the transcoding:

Among Fig. 2, the angle of θ presentation video texture, (u v) is the DCT coefficient to F.Wherein the DCT coefficient is the DCT variation coefficient in each macro block in the MPEG input code flow, by calculating θ, and the intra prediction mode of selection and θ equidirectional, can obtain upper right, level and vertical frame inner estimation mode.Yet the θ that adopts this method calculating gained can't obtain all prediction direction (as the bottom right prediction) of AVS intra prediction mode in the scope of 0～90 degree.Therefore, further introduce parameter ρ, be used for the angular range of auxiliary judgment θ, parameter ρ is definite by the sign symbol of the data of [0] [1] of 8 * 8DCT coefficient matrix and position, [1] [0], the lowest frequency component that these two coefficients are respectively the AC coefficients, reflected the basic texture information of image:

The texture that DCT coefficient [0] [1] position parameter is represented is:

Just

Negative

The texture that DCT coefficient [1] [0] position parameter is represented is:

Just

Negative

[0] when [1], [1] [0] two positions coefficient jack per line image roughly texture be:

Just

Negative

[0] when [1], [1] [0] two positions coefficient contrary sign image roughly texture be:

Positive and negative

Just bear

As the above analysis, when two coefficient jack per lines (ρ=1) image texture be roughly the upper right side to, image texture is roughly the lower right to, incorporating parametric θ during contrary sign (ρ=0), grain direction that can the predicted image data piece.

In intra prediction mode is judged, calculate θ and ρ can obtain current block to be encoded with and a left side is adjacent and the grain direction of last adjacent data block, when block to be encoded with left side neighbour or when upward adjacent data block grain direction is consistent, can obtain intra prediction mode fast.When grain direction exists than big-difference, then adopt and further calculate prediction matrix in the DCT territory, relatively the SAD of intra prediction mode and current data block obtains the optimum frame inner estimation mode, and concrete algorithm process flow process is as follows:

1. judge whether current block to be encoded is in image left hand edge or top edge,, otherwise enter step 2 if then be judged as the DCT intra prediction mode;

2. whether judgement current macro, the θ value of left adjacent piece and last adjacent piece all satisfy T1＜θ＜T2 (T1=35 °, T2=55 °), if illustrate that the current macro region has certain oblique texture, carry out step 3, otherwise carry out step 4;

3. judge the value of ρ, if ρ=1 then can judge that the intra prediction mode of current block to be encoded is upper right intra prediction mode, ρ=0 is judged as the bottom right intra prediction mode;

4. whether the θ value of judging current macro and last adjacent piece all satisfies 0 °＜=θ＜T3 (T3=20 °), if illustrate that the current macro region has certain vertical texture, can be judged as the vertical frame inner estimation mode, otherwise carry out step 5;

5. whether the θ value of judging current macro and left adjacent piece all satisfies T4＜θ＜=90 ° (T4=70 °), if illustrate that the current macro region has the certain level texture, can be judged as horizontal intra prediction mode, otherwise carry out step 6;

6. the DCT prediction data (computational methods see next section) of calculated level, vertical and three kinds of patterns of DC, and calculate respectively its with the DCT data absolute error of current block to be encoded with:

$\mathrm{SAD}=|\mathrm{DCT}\left(D\right)-\mathrm{DCT}\left(S\right)|=\underset{i=0}{\overset{i<8}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{j<8}{\mathrm{\&Sigma;}}}|D_{\mathrm{ij}}-S_{\mathrm{ij}}|---\left(0.1\right)$

7. remember SAD _HZFor the absolute error of horizontal forecast and, SAD _VTFor the absolute error of vertical prediction and, SAD _DCFor the absolute error of DC prediction and.By comparing SAD _HZ, SAD _VTAnd SAD _DC, determine that the pairing pattern of reckling is the optimum frame inner estimation mode.

8. by above-mentioned algorithm flow as seen, in most of the cases, need not carry out the calculating of the 6th step and can obtain intra prediction mode.The 6th step handled the prediction matrix that relates to the DCT territory and calculates.Represent dct transform with DCT (●) in below discussing, D represents 8 * 8 pixel initial data, S remarked pixel territory prediction data, the extraction matrix of R remarked pixel territory prediction data, P represents the extraction matrix of DCT territory prediction data, multiplies each other and just can obtain the prediction matrix of correspondence with extracting matrix and raw data matrix in pixel domain and DCT territory.Y represents the pixel domain weighting matrix in the DC intra prediction mode, and Q represents the DCT territory weighting matrix in the DC intra prediction mode.Different subscripts has been represented various intra prediction modes." * " expression multiplication of determinants,

The point multiplication operation of representing matrix correspondence position.Be the computational methods of example explanation horizontal forecast and vertical prediction DCT prediction matrix below with the vertical prediction.In pixel domain, vertical prediction matrix data S _UpThe employing following formula draws:

S _up＝R _up×D _up (0.2)

$S_{\mathrm{up}}=\left[\begin{array}{cccccccc}{d}_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\end{array}\right];$ $R_{\mathrm{up}}=\left[\begin{array}{cccccccc}0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\end{array}\right];$ $D_{\mathrm{up}}=\left[\begin{array}{cccccccc}{d}_{00}& d_{01}& d_{02}& d_{03}& d_{04}& d_{05}& d_{06}& d_{07}\\ d_{10}& d_{11}& d_{12}& d_{13}& d_{14}& d_{15}& d_{16}& d_{17}\\ d_{20}& d_{21}& d_{22}& d_{23}& d_{24}& d_{25}& d_{26}& d_{27}\\ d_{30}& d_{31}& d_{32}& d_{33}& d_{34}& d_{35}& d_{36}& d_{37}\\ d_{40}& d_{41}& d_{42}& d_{43}& d_{44}& d_{45}& d_{46}& d_{47}\\ d_{50}& d_{51}& d_{52}& d_{53}& d_{54}& d_{55}& d_{56}& d_{57}\\ d_{60}& d_{61}& d_{62}& d_{63}& d_{64}& d_{65}& d_{66}& d_{67}\\ d_{70}& d_{71}& d_{72}& d_{73}& d_{74}& d_{75}& d_{76}& d_{77}\end{array}\right]---\left(0.3\right)$

Wherein: D _Up8 * 8 data blocks of representing current block to be encoded top, d _IjBe the pixel data of correspondence position wherein.

Under the condition of not considering computational accuracy, dct transform is linear transformation.Formula (0.2) is carried out dct transform, has:

DCT(S _up)＝DCT(R _up×D _up)

＝DCT(R _up)×DCT(D _up) (0.4)

＝P _up×DCT(D _up)

As seen, the DCT territory prediction data of vertical frame inner estimation mode can be by the dct transform data (DCT (D of pixel data _Up)) and the extraction matrix (P of DCT territory prediction data _Up) obtain.P wherein _UpCan pass through R _UpCarrying out dct transform tries to achieve:

$P_{\mathrm{up}}=\mathrm{DCT}\left(R_{\mathrm{up}}\right)=\mathrm{DCT}\left(\left[\begin{array}{cccccccc}0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0& 0& 1\end{array}\right]\right)=\left[\begin{array}{cccccccc}1.0003& -1.3872& 1.3066& -1.1759& 1.0003& -0.7858& 0.5411& -0.2758\\ 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0& 0& 0\end{array}\right]---\left(0.5\right)$

Adopt above-mentioned similar approach can obtain the DCT territory prediction matrix of horizontal forecast:

DCT(S _left)＝DCT(D _left)×P _left (0.6)

Wherein: P _Left=P _Up ^T, T representing matrix transposition.

The prediction matrix that needs in the DCT intra prediction mode horizontal intra prediction mode or vertical frame inner estimation mode are obtained is done a weighted filtering and is handled, and concrete weighting scheme is referring to standard.With the available DC intra prediction mode of left piece only is that example illustrates the matrix processing method of this weighting procedure in the DCT territory, and the generation of DCT prediction matrix in the DC intra prediction mode, in pixel domain:

S _{dc_left}＝(Y _left/4)×S _left $Y_{\mathrm{left}}=\left[\begin{array}{cccccccc}3& 1& 0& 0& 0& 0& 0& 0\\ 1& 2& 1& 0& 0& 0& 0& 0\\ 0& 1& 2& 1& 0& 0& 0& 0\\ 0& 0& 1& 2& 1& 0& 0& 0\\ 0& 0& 0& 1& 2& 1& 0& 0\\ 0& 0& 0& 0& 1& 2& 1& 0\\ 0& 0& 0& 0& 0& 1& 2& 1\\ 0& 0& 0& 0& 0& 0& 1& 3\end{array}\right]---\left(0.7\right)$

S wherein _{Dc_left}But the pixel domain prediction matrix of representing only left piece time spent DC intra prediction mode, S _LeftThe pixel domain prediction matrix of representing horizontal intra prediction mode, Y _LeftThe weighting matrix in remarked pixel territory.Formula (0.7) is carried out dct transform, has:

DCT(S _{dc_left})＝DCT((Y _left/4)×S _left) (0.8)

Can get by mathematics arrangement (0.7) formula:

$\mathrm{DCT}\left(S_{\mathrm{dc}\_\mathrm{left}}\right)=Q_{\mathrm{left}}\&CircleTimes;\mathrm{DCT}\left(S_{\mathrm{left}}\right)$ $Q_{\mathrm{left}}=\left[\begin{array}{cccccccc}1.0000& 0& 0& 0& 0& 0& 0& 0\\ 0.9619& 0& 0& 0& 0& 0& 0& 0\\ 0.8535& 0& 0& 0& 0& 0& 0& 0\\ 0.6914& 0& 0& 0& 0& 0& 0& 0\\ 0.5000& 0& 0& 0& 0& 0& 0& 0\\ 0.3087& 0& 0& 0& 0& 0& 0& 0\\ 0.1464& 0& 0& 0& 0& 0& 0& 0\\ 0.0381& 0& 0& 0& 0& 0& 0& 0\end{array}\right]---\left(0.9\right)$

By formula (0.8) as seen, the available DCT prediction data of only left piece can be by the DCT prediction data (DCT (S of horizontal intra prediction mode _Left)) and weighting matrices Q _LeftObtain.Adopt above-mentioned similar approach can obtain only going up the available DCT territory prediction data of piece:

$\mathrm{DCT}\left(S_{\mathrm{dc}\_\mathrm{up}}\right)=Q_{\mathrm{up}}\&CircleTimes;\mathrm{DCT}\left(S_{\mathrm{up}}\right)---\left(0.10\right)$

But further can obtain a left side, go up all DCT prediction data of time spent of blocks of data:

DCT(S _{dc_both})＝(DCT(S _{dc_left})+DCT(S _{dc_up}))/2 (0.11)

Q wherein _Up=Q _Left ^T

In the said process, DCT (D) data can the decoded portion in transcoding directly obtain, and D then selects the adjacent or upward adjacent data block in a left side according to different mode.Adopt extraction matrix P and DCT (D) to multiply each other, can obtain the pairing DCT prediction data of level and vertical mode DCT (S).Further use weighting matrices Q and DCT (S) to carry out point multiplication operation, can obtain DCT and predict pairing DCT prediction data.Most of data are 0 among P and the Q, and are comparatively simple when therefore calculating.

Two, motion vector is reused and the inter macroblock evaluation algorithm

1, the inter macroblocks partition mode is judged

The inter macroblocks of MPEG is divided in only supports 16 * 16 1 kinds of coding modes under the frame encoding mode, increased by 16 * 8 coding modes under the coding mode on the scene.AVS supports 16 * 16,16 * 8 on inter macroblock, 8 * 16,8 * 8 four kinds of coding modes.Therefore at MPEG in AVS transform coding process, solve the problem that rejudges of macro-block partition mode with regard to needs.

For most of natural sequence, the probability of use maximum of 16 * 16 macro-block partition modes, other pattern proportion is less.MPEG of the present invention is in the real-time transcriber design of AVS, and the inter macroblocks partition mode is selected 16 * 16 and 8 * 8 two kinds.Do not use the main cause of 16 * 8 and 8 * 16 two kinds of macro-block partition modes that 2 points are arranged: the first, consider from amount of calculation, reduce this two patterns, judge that in inter macroblock the amount of calculation of link has just reduced about 50%, help reaching the requirement of real-time transcoding; The second, 16 * 8 and 8 * 16 these two kinds of patterns also are to use than multizone in details, and the more situation of more and vertical texture at cross grain can be finished under the situation of only bringing a motion vector expense for both of these case 8 * 8 patterns respectively.

Inter macroblocks partition mode judgement motion vector in Fig. 1 is reused with the inter macroblock judge module and is finished, and this module mainly adopts the macro block level of detail as judgment criterion when judging macro-block partition mode, and concrete deterministic process is:

The first step, all inter macroblocks all are initialized as 16 * 16 macro-block partition modes;

Second step, the VLC number of coded bits of each the macro block residual error coefficient in the statistics MPEG code stream, calculate the VLC coding average number of bits of present frame macro block residual error coefficient, and whether belong to the basic thresholding of details as macro block than the judgment criteria of multizone with this average number of bits;

The 3rd step, be divided by with the VLC number of coded bits and the basic thresholding of each macro block residual error coefficient, if ratio greater than 1.5 times (experiment setting thresholds), the level of detail that this macro block then is described needs use 8 * 8 patterns to encode obviously greater than the image averaging value; Otherwise use 16 * 16 patterns of acquiescence.

2, inter macroblock evaluation algorithm

For inter macroblocks, the AVS standard provides multiple type of coding.Though the MPEG macroblock encoding type can not be corresponding one by one with the AVS macroblock encoding type, can not directly be used at coding side, it but can reflect the region characteristic of image, thereby judges AVS macroblock encoding type according to these region characteristics.Table 1, table 2 have provided in the transform coding process between mpeg frame macroblock encoding type to the conversion initialization mapping table of AVS inter macroblocks type of coding.

Table 1 P frame macroblock encoding type initialization mapping table

Table 2 B frame macroblock encoding type initialization mapping table

48 * 8 of B frame macro block pieces adopt different prediction direction to encode among the AVS, in order to obtain the type of coding of 8 * 8 sub-pieces, need further handle, as Fig. 3, the type of coding of initialization 8 * 8 sub-piece E is the mpeg encoded type of the sub-piece of E place macro block, current coding macro block is a forward predictive coded among the MPEG, and then the sub-initialization block of E is SB_Fwd_8 * 8; Obtain the mpeg encoded type of the sub-piece adjacent macroblocks of E A, B, C: if the initialization codes type of the sub-piece of E is SB_Fwd_8 * 8 or SB_Sym_8 * 8, and the MPEG macroblock encoding type of A, B, C three macro blocks is the back forecast coding, the type of coding of then judging the sub-piece of E is SB_Bck_8 * 8, otherwise is judged as the initialization codes type of E; If the initialization codes type of the sub-piece of E is SB_Bck_8 * 8 or SB_Sym_8 * 8, and the MPEG macroblock encoding type of A, B, C three macro blocks is forward predictive coded, the type of coding of then judging the sub-piece of E is SB_Fwd_8 * 8, otherwise is judged as the initialization codes type of E.

The step of macro block mode evaluation algorithm is as follows:

Step 1, if the mpeg decode macro block corresponding with current coding macro block is intraframe predictive coding, then this macro block adopts the AVS intraframe predictive coding, because around the current macro all is inter macroblocks, has only residual error data, so can not use DCT territory intra prediction mode to judge as prosthomere, directly select the DC pattern here;

Step 2, the division size of judgement inter macroblocks: if 16 * 16 division sizes then carry out steps 3; If 8 * 8 division sizes then carry out steps 4;

Step 3, the type of coding that carries out under 16 * 16 partition modes according to table 1, table 2 shines upon;

Step 4, carry out the type of coding initialization mapping of 8 * 8 partition modes according to table 1, table 2, obtain the final type of coding of all 8 * 8 sub-pieces in the current coding macro block then according to 8 * 8 subblock coding type bearing calibrations of introducing previously, and encode expense relatively with 16 * 16 partition modes, select reckling to encode;

Step 5 is proofreaied and correct if B frame macro block then carries out the Direct pattern;

Step 6 is carried out intraframe predictive coding to the inter macroblocks that satisfies infra-frame prediction macroblock coding correcting condition and is proofreaied and correct, and the intra prediction mode decision method is with step 1.

3, motion vector is reused and optimized Algorithm

Reusing of motion vector is key technology in the video transformation coding.In the transform coding process between identical standard, the motion vector of input code flow is through the encoder motion estimation process, and it is representing the optimal motion vector under certain motion estimation algorithm.Transform coding between various criterion, because differences such as reference picture interpolation process and motion vector precision, motion vector is reused the immediate interest in the time of can't bringing the similar code stream of picture to change, the important references motion vector when still but can be used as estimation.The brightness reference picture of MPEG is a half-pixel accuracy, and its half-pix point is to come by two adjacent whole pixel samples linear interpolations.The brightness reference picture of AVS is 1/4 pixel precision, and its interpolation process is referring to normative document.Compare the computational process of half-pix point in these two standards, MPEG adopts linear interpolation filter, 4 tap filters (1/8 that AVS adopts, 5/8,5/8 ,-1/8), though these two kinds of half-pix interpolation filter structure differences, final interpolation result be reflected on the image difference not clearly.Therefore, can think that the half-pixel accuracy motion vector of MPEG is similar to the optimum half-pixel accuracy motion vector that AVS encoder motion search obtains, during transcoding with the half-pixel accuracy motion vector (in Fig. 4 G be depicted as the reference block locations of this half-pixel accuracy motion vector indication) of mpeg decode motion vector as optimum, thereby whole pixel motion vector search and half-pix motion-vector search process have been saved, the motion vector of doing one time 1/4 pixel precision on this motion vector basis is heavily searched for optimization then, promptly search for this half-pix motion vector 8 1/4 pixel precision motion vectors (a～h is the reference block locations of 8 1/4 pixel precision motion vector indications among Fig. 4) on every side, thereby obtain the motion vector of optimum 1/4 pixel precision.

Coding mode for adopting 16 * 16 directly adopts above method, obtains optimum 1/4 pixel precision motion vector; For 8 * 8 macro-block partition modes, because MPEG has only the motion vector of macro-block level, therefore need each 8 * 8 sub-piece of absolute coding, each 8 * 8 sub-piece all adopts the macro-block level motion vector of MPEG, and then does heavily search respectively and optimize.

Below by a concrete embodiment, further specify technical scheme of the present invention.

Experiment adopts reference software rm52j_r1 that AVS standard operation group provides as a reference, in the experimental result with representing among the REF.The MPEG that this patent design realizes represents with TRANS in experimental result to the AVS trans-coding system.The experiment sequence is for using bus_cif, mobile_cif, football_704 * 480, the MPEG PS stream (frame per second: 30) that tempete_704 * 480 standard test sequences are made, and be used for actual verification DVD program (MPEG PS stream) " time " (720 * 576, frame per second: 30).Experimental situation is P41.5GHZ, the PC of 320M internal memory.Each sequence R-D curve as shown in Figure 5 and Figure 6.

The running time of reference encoder device REF and this paper trans-coding system TRANS during table 3 has provided and tested, PSNR and bit rate, and the comparative result of amount of calculation (time ratio inverse), quantization parameter is QP=35.

The comparison of the time of the different cycle testss of table 3, PSNR and bit rate

Fig. 5,6 and the experimental result of table 3 show that the trans-coding system that this patent proposes is compared under same code rate with the reference encoder device, image PSNR slightly reduces.The most obvious in the reduction of football sequence, reason is because this sequence motion is very fast and complicated, and the motion vector of mpeg encoded is bigger, has increased AVS coding expense when transcoding, causes the bit rate rising.As shown in Table 3, this patent trans-coding system is under the prerequisite that guarantees output image quality, and amount of calculation has only about 15% of reference encoder device, has also adopted other cycle testss to obtain consistent result in the experiment.

In order to verify the correctness of transcoding code stream, the AVS Standard Digital Television set-top box that has also adopted Co., Ltd of Shanghai Guangdian Electric Group in the experiment is as the aided verification means.The AVS transport stream (AVS TS) that transcoding is obtained sends to set-top box by code stream transmission integrated circuit board, and television set is sent in the audio/video output of set-top box, transport stream playing programs smoothness, and audio-visual synchronization is good.

Claims (5)

1. a MPEG is to the video coding conversion system of AVS, the data flow, mpeg system layer demultiplexing module and the mpeg system layer Multiplexing module that comprise mpeg format, it is characterized in that, this system also comprises coded conversion device, and the DCT intra prediction mode judge module in this coder transitions, motion vector is reused and the inter macroblock judge module, and wherein motion vector is reused and the inter macroblock judge module is divided into that motion vector is reused submodule and inter macroblock is judged submodule; The media stream of mpeg format at first passes through mpeg system layer demultiplexing module, extracts MPEG video, voice data and other data behind the demultiplexing respectively; Wherein the mpeg audio data that obtain of demultiplexing are not carried out transcoding and are handled, directly send into " the mpeg system layer is multiplexing " module of rear end, and the MPEG video data that coder transitions is exported demultiplexing is fully decoded to pixel domain, again in conjunction with DCT intra prediction mode judge module, motion vector is reused and the inter macroblock judge module carries out AVS coding, sends into mpeg system layer Multiplexing module then; DCT intra prediction mode judge module in the transducer utilizes decode again DCT coefficient behind the inverse quantization of the VLC of input code flow to carry out intra prediction mode and judge; Motion vector reuse and the inter macroblock judge module in motion vector reuse submodule with the reference of the motion vector MV in the MPEG code stream as the AVS estimation, after motion vector MV optimization, obtain half-pixel accuracy motion vector MV, inter macroblock judges that submodule is then by the mpeg decode macro block, judge AVS macroblock encoding type, divide size.

2. MPEG as claimed in claim 1 is characterized in that to the video coding conversion system of AVS the algorithm of described DCT intra prediction mode judge module may further comprise the steps:

In intra prediction mode is judged, the angle θ of computed image texture and parameter ρ, obtain current block to be encoded with and a left side is adjacent and the grain direction of last adjacent data block, when macroblock to be encoded with left side neighbour or when upward adjacent data macro grain direction is consistent, obtain intra prediction mode

$\mathrm{\&theta;}=\arctan \left(\frac{\underset{u=1}{\overset{7}{\mathrm{\&Sigma;}}}\left|F(u,0)\right|}{\underset{v=1}{\overset{7}{\mathrm{\&Sigma;}}}\left|F(0,v)\right|}\right)$

In the formula, F (u, 0) is the first row AC coefficient of DCT coefficient; F (0, v) be the first row AC coefficient of DCT coefficient;

When the adjacent or upward adjacent data macro grain direction in a macroblock to be encoded and a left side exists than big-difference, be that macroblock to be encoded, a left side are adjacent and go up and have two to satisfy 20 °＜=θ＜=35 ° or 55 °＜=θ＜=70 ° respectively and can not carry out level, vertical and oblique prediction the time in the adjacent macro block, need further to calculate prediction matrix in the DCT territory, the DCT data absolute error and the SAD that compare the current block to be encoded of intra prediction mode and current data block, obtain the optimum frame inner estimation mode, concrete algorithm process flow process is as follows:

Step 1 judges whether current block to be encoded is in image left hand edge or top edge, if then be judged as the DCT intra prediction mode, otherwise enter step 2;

Step 2 is judged current macro, and whether the θ value of left adjacent piece and last adjacent piece all satisfies T1＜θ＜T2, and T1=35 °, T2=55 ° are if illustrate that the current macro region has certain oblique texture, carry out step 3, otherwise carry out step 4;

Step 3 is judged the value of parameter ρ, if ρ=1 then can judge that the intra prediction mode of current block to be encoded is upper right intra prediction mode, ρ=0 is judged as the bottom right intra prediction mode;

Step 4 judges whether the θ value of current macro and last adjacent piece all satisfies 0 °＜=θ＜T3 (T3=20 °), if illustrate that the current macro region has certain vertical texture, can be judged as the vertical frame inner estimation mode, otherwise carry out step 5:

Whether step 5, the θ value of judging current macro and left adjacent piece all satisfy T4＜θ＜=90 °, and T4=70 °, if illustrate that the current macro region has the certain level texture, can be judged as horizontal intra prediction mode, otherwise carry out step 6;

Step 6, calculate respectively itself and current block to be encoded DCT data absolute error and:

$\mathrm{SAD}=|\mathrm{DCT}\left(D\right)-\mathrm{DCT}\left(S\right)|=\underset{i=0}{\overset{i<8}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{j<8}{\mathrm{\&Sigma;}}}|D_{\mathrm{ij}}-S_{\mathrm{ij}}|$

Step 7, note SAD _HZFor the absolute error of horizontal forecast and, SAD _VTFor the absolute error of vertical prediction and, SAD _DCFor the absolute error of DC prediction and, by SAD relatively _HZ, SAD _VTAnd SAD _DC, determine that the pairing pattern of reckling is the optimum frame inner estimation mode;

Above-mentioned parameter parameter ρ determines that by the sign symbol of the data of [0] [1] of 8 * 8DCT coefficient matrix and position, [1] [0] when two coefficient jack per lines, parameter ρ=1 is when two coefficient contrary signs, parameter ρ=0.

3. MPEG as claimed in claim 1 is to the video coding conversion system of AVS, it is characterized in that, described motion vector is reused and the inter macroblock judge module, this module mainly adopts the macro block level of detail as judgment criterion when judging macro-block partition mode, the inter macroblocks partition mode is selected 16 * 16 and 8 * 8 two kinds, and concrete deterministic process may further comprise the steps:

Step 1, all inter macroblocks all are initialized as 16 * 16 macro-block partition modes;

Step 2, the VLC number of coded bits of each the macro block residual error coefficient in the statistics MPEG code stream, calculate the VLC coding average number of bits of present frame macro block residual error coefficient, and whether belong to the basic thresholding of details as macro block than the judgment criteria of multizone with this average number of bits;

Step 3 is divided by with the VLC number of coded bits and the basic thresholding of each macro block residual error coefficient, if ratio greater than 1.5 times (experiment setting thresholds), the level of detail that this macro block then is described needs to use 8 * 8 patterns to encode obviously greater than the image averaging value; Otherwise use 16 * 16 patterns of acquiescence.

4. MPEG as claimed in claim 1 is characterized in that to the video coding conversion system of AVS, described motion vector reuse and the inter macroblock judge module in inter macroblock judge and may further comprise the steps the algorithm of submodule:

Judge AVS macroblock encoding type according to region characteristic, table 1, table 2 be in the transform coding process between mpeg frame macroblock encoding type to the conversion initialization mapping table of AVS inter macroblocks type of coding;

Table 1P frame macroblock encoding type initialization mapping table

Table 2B frame macroblock encoding type initialization mapping table

48 * 8 of B frame macro block pieces adopt different prediction direction to encode among the AVS, in order to obtain the type of coding of 8 * 8 sub-pieces, need further handle:

The type of coding of initialization 8 * 8 sub-piece E is the mpeg encoded type of the sub-piece of E place macro block, and current coding macro block is a forward predictive coded among the MPEG, and then the sub-initialization block of E is SB_Fwd_8 * 8; Obtain the mpeg encoded type of the sub-piece adjacent macroblocks of E A, B, C: if the initialization codes type of the sub-piece of E is SB_Fwd_8 * 8 or SB_Sym_8 * 8, and the MPEG macroblock encoding type of A, B, C three macro blocks is the back forecast coding, the type of coding of then judging the sub-piece of E is SB_Bck_8 * 8, otherwise is judged as the initialization codes type of E; If the initialization codes type of the sub-piece of E is SB_Bck_8 * 8 or SB_Sym_8 * 8, and the MPEG macroblock encoding type of A, B, C three macro blocks is forward predictive coded, the type of coding of then judging the sub-piece of E is SB_Fwd_8 * 8, otherwise is judged as the initialization codes type of E;

The step of inter macroblock evaluation algorithm is as follows:

Step 1, if the mpeg decode macro block corresponding with current coding macro block is intraframe predictive coding, then this macro block adopts the AVS intraframe predictive coding, because around the current macro all is inter macroblocks, has only residual error data, so can not use DCT territory intra prediction mode to judge as prosthomere, directly select the DC pattern here;

Step 2, the division size of judgement inter macroblocks: if 16 * 16 division sizes then carry out steps 3; If 8 * 8 division sizes then carry out steps 4;

Step 3, the type of coding that carries out under 16 * 16 partition modes according to table 1, table 2 shines upon;

Step 4, carry out the type of coding initialization mapping of 8 * 8 partition modes according to table 1, table 2, obtain the final type of coding of all 8 * 8 sub-pieces in the current coding macro block then according to 8 * 8 subblock coding type bearing calibrations of introducing previously, and encode expense relatively with 16 * 16 partition modes, select reckling to encode;

Step 5 is proofreaied and correct if B frame macro block then carries out the Direct pattern;

Step 6 is carried out intraframe predictive coding to the inter macroblocks that satisfies infra-frame prediction macroblock coding correcting condition and is proofreaied and correct, and the intra prediction mode decision method is with step 1.

5. MPEG as claimed in claim 1 is characterized in that to the video coding conversion system of AVS, described motion vector reuse and the inter macroblock judge module in motion vector reuse the algorithm of submodule, may further comprise the steps:

For adopting 16 * 16 coding mode and adopting 8 * 8 macro-block partition mode, all with the half-pixel accuracy motion vector of mpeg decode motion vector as optimum, the motion vector of doing one time 1/4 pixel precision on this motion vector basis is heavily searched for optimization then, promptly search for this half-pix motion vector 8 1/4 pixel precision motion vectors on every side, thereby obtain the motion vector of optimum 1/4 pixel precision, and 4 motion vectors of macro-block partition mode record that adopt 8 * 8.

Images