CN101170688B - A quick selection method for macro block mode - Google Patents

Abstract

The invention relates to a rapid selection method of macro-block mode during video information processing procedure, particularly a rapid selection method suitable for P-slice macro-block mode. The method comprises the following steps: step 1, determining the size of continuous macro-block according to the size of on-chip memory in the beginning, dividing all macro-blocks in one frame into a plurality of continuous 16is multiplied by16 macro-blocks, all macro-blocks in each continuous macro-block being positioned in the same row in the same one frame, and distributing the desired on-chip buffer zone; step 2, copying the brightness and colority data required for the mode selection of continuous macro-blocks in the current frame and the reference frame to the on-chip buffer zone; step 3, calculating the mode of each macro-block in the continuous macro-blocks to determine the final mode of each macro-block; and step 4, if all macro-blocks in the current continuous macro-block are processed, stopping the selection procedure of the current continuous macro-block, otherwise returning to the step 1 to continue. The inventive technical proposal can efficiently accelerate the speed of video compression and calculation, thereby realizing the real-time compression of video.

Description

A kind of fast selecting method of macro block mode

Technical field

The present invention relates to the fast selecting method of macro block mode in a kind of video information process process, particularly a kind of fast selecting method that is applicable to the macro block mode of P-slice.

Background technology

AVS (Audio-Video Standard) digital audio/video encoding and decoding technique standard is the common base technical standard of digital audio/video industries such as China's Digital Television, high definition optic disk, Streaming Media and multimedia, has represented current international most advanced level.H.264 the video compression coding standard of new generation formulated of the joint video team of forming by ISO/IEC and ITU-T (JVT), based on video compression standard in advance, H.264 having introduced a lot of advanced persons' technology, also is one of present internationally recognized advanced video technical standard.

P-slice is AVS, H.264 waits in the standard a kind of slice the most commonly used, and AVS, other video encoding and decoding standard that H.264 reaches extensive use in recent years all are that to adopt with macro block (macro-block) be the code encoding/decoding mode of base unit to P-slice.From the prior art, selection for macro block mode mainly contains following certain methods: (1) adopts rate-distortion optimization (RDO:Rate-DistortionOptimization) method to calculate the rate distortion costs (RDCost:Rate-Distortion Cost) of various macro block modes, and the pattern of rate distortion costs minimum is used as final coding mode, this method has improved code efficiency and picture quality, but it needs that each possible macro block mode is carried out rate-distortion optimization and calculates, and makes amount of calculation sharply increase.(2) according to motion search calculation cost functional value, the sort of macro block mode of selected value minimum is as final pattern, this method can be good the decoded picture quality of assurance, but amount of calculation is also very big, makes estimation become one of the most time-consuming part of whole video coding.(3) division of adopting the level of detail of macroblock image to come the macro block of Inter pattern is carried out sub-piece, thereby the final pattern (the sub-block type of inter macro block) of decision inter macro block, the amount of calculation of this method is little, but be not easy on intraframe coding (intra) or interframe encode (inter) pattern, make one's options, thereby can not select the macro block mode of p-slice effectively.Is (immediate prior art the 3rd kind?)

Summary of the invention

In order to accelerate the video compression computational speed, realize the real-time video compression, the invention provides a kind of fast selecting method at the P-slice macro block mode.

Technical scheme of the present invention is: a kind of fast selecting method of macro block mode, the method comprising the steps of: step 1, determine the size of continuous macro block during beginning according to the size of on-chip memory, is all macroblock partitions in the frame 16 * 16 macro blocks of plurality of continuous, all macro blocks in each continuous macro block all are in the frame with in the delegation, and distribute required sheet internal buffer; Step 2, in present frame and the reference frame for continuous macro block carry out the needed brightness of model selection, chroma data copies to the sheet internal buffer; Step 3 is calculated to determine the final pattern of each macro block the pattern of each macro block in the continuous macro block; Step 4 if all macro block modes are handled in the current continuous macro block, then finishes current continuous Macroblock Mode Selection process; Otherwise getting back to step 1 proceeds.

Further, in step 3, the current macro in each continuous macro block is carried out the whole pixel motion estimation of 16 * 16 patterns, and carry out continuous circular treatment for the calculating of all macro block modes in each continuous macro block, macro block is not carried out operations such as conversion, quantification, reconstruction, variable-length encoding therebetween.

The invention has the beneficial effects as follows: owing to be all macroblock partitions in the frame plurality of continuous macro block, all macro blocks in each continuous macro block all are in the frame with in the delegation, and the needed brightness of model selection, chroma data copied in the sheet internal buffer, reduced the exchange of data, make the speeding up of data access, accelerated the computational speed of video compression.When preference pattern, do not need macro block is carried out operations such as conversion, quantification, reconstruction, variable-length encoding, make computation complexity greatly reduce, further accelerated the computational speed of video compression.

Description of drawings

Fig. 1 shows the flow chart according to the fast selecting method of a kind of macro block mode of the embodiment of the invention;

Fig. 2 shows the numbering of 48 * 8 sub-pieces of brightness.

Embodiment

Below in conjunction with drawings and Examples is that example further specifies the present invention:

In the AVS standard, macro block mode is divided into interframe encode (inter) pattern and intraframe coding (intra) pattern, interframe encoding mode can be divided into 16 * 16,16 * 8,8 * 16,8 * 8 patterns according to sub-piece size again, and the skip mode among the P-slice is a kind of special pattern of interframe encoding mode.The sub-piece of each 8 * 8 brightness comprises 5 kinds of predictive modes in the intra-frame encoding mode, and the sub-piece of each 8 * 8 colourity comprises 4 kinds of predictive modes.Encoder comes the pattern of macro block is selected by motion estimated values, degrees of detail being worth isoparametric calculating.

In the present embodiment, select the hardware platform of the TMS320DM64x series multimedia process chip of company of Texas Instruments (TI) for use as present embodiment.The TMS320DM64x series multimedia process chip of company of Texas Instrument (TI), be suitable for digital media applications, have level cache (cache) and configurable RAM/Cache on the DM64x sheet, and 64 32 general registers, these memory spaces have guaranteed the efficiently operation apace of extensive audio frequency and video handling procedure.Level cache is divided into two kinds of program cache and data cache, and access speed is fast.Configurable RAM/Cache can be divided into on-chip memory (ISRAM) and L2 cache to it, and on-chip memory (ISRAM) can be deposited program and data, and access speed is fast.L2 cache also is a kind of high speed cache, can improve the access speed of program and data.The applied EDMA controller of DM64x has 64 independent DMA passages, and its transmission of configuration information is kept among the RAM, can handle a plurality of DMA transformation tasks simultaneously.The DMA transmission only needs DSP nuclear minimum degree ground to get involved, thereby has improved the processing speed of system effectively.Select for use the P-SLICE of AVS standard to carry out model selection in the present embodiment.

Flow chart has as shown in Figure 1 provided the detailed process of present embodiment:

Step 1 is determined the size of continuous macro block according to the size of the effective on-chip memory of DSP (ISRAM) when initial, and is distributed required sheet internal buffer etc.The size (the macro block number that it comprises) of continuous macro block is the span decision according to the size of the ISRAM of energy usefulness and motion vector:

Usually, the size of establishing continuous macro block is L, the span of motion vector is-and N～+ N, the size of effective I SRAM is M ₀Byte, reference frame number are 1 frame, then

$L=\left[\frac{M_0-416-64N-{8\mathrm{<figure-callout\; id="2"\; label="N"\; filenames="S2007100506189E00021.png"\; state="\{\{state\}\}">N</figure-callout>}}^2}{928+64N}\right]$

Thereby the big or small L according to the continuous macro block that calculates determines actual required sheet internal buffer size M (byte),

M＝2×(16×L+2N)×(16+2N)+(L+1)×16×26

Step 2, in present frame and the reference frame for continuous macro block carry out the needed brightness of model selection, chroma data copies to the sheet internal buffer by the mode of DMA.These data comprise: brightness, chroma data in the continuous macro block of present frame, and the lastrow data and the left data of continuous macroblock boundaries; The brightness of correspondence position, chroma data in the reference frame.

Step 3, the mode computation of each macro block in the macro block continuously.

Step 3.1 is carried out the whole pixel motion of 16 * 16 patterns and is estimated to current macro in the continuous macro block.

Determine the position of current macro in the continuous macro block, and use the method for our unit's patent 1 to carry out the fast motion estimation calculating of 16 * 16 block modes, establishing the motion vector that obtains after estimation is finished is MV ₁=(MVX ₁, MVY ₁), residual values is ${\mathrm{<figure-callout\; id="1"\; label="SAD"\; filenames="S2007100506189E00021.png"\; state="\{\{state\}\}">SAD</figure-callout>}}_1=\underset{i,j}{\mathrm{\&Sigma;}}|X_{i,j}-Y_{i,j}|.$ X wherein _{I, j}And Y _{I, j}Be respectively (i, j) pixel value of position in present frame and the reference frame.

Does step 3.2 judge that according to motion estimation result the originate mode of current macro is the Intra pattern?

Determine a threshold values T ₀=16 * (max (min (40, q), 24)) ², wherein q is a quantization parameter.Work as SAD ₁＞T ₀The time, the originate mode of current macro as the Intra pattern, is proceeded to step 3.8; Otherwise, proceed to next step 3.3.

Does step 3.3 judge that current macro is the SKIP pattern?

Determine a threshold values T ₁=256 * (max (min (6, q/6), 4)) ², wherein q is a quantization parameter.Work as SAD ₁＜T ₁, and MV ₁When equaling the predicted value of motion vector, the originate mode of current macro as the SKIP pattern, is proceeded to step 3.8; Otherwise, proceed to next step 3.4.

Are step 3.4, the originate mode of current macro 16 * 16?

Determine a threshold values T ₂=768 * (max (min (16, q/3), 8)), wherein q is a quantization parameter.Work as SAD ₁＜T ₂, the originate mode of current macro as 16 * 16 patterns, is proceeded to step 3.8; Otherwise, proceed to next step 3.5.

Step 3.5, the degrees of detail of 48 * 8 sub-pieces of brightness is determined its initial inter pattern in the calculating current macro.

The numbering of 48 * 8 sub-pieces of brightness is illustrated in fig. 2 shown below.

For each 8 * 8 sub-piece, at first calculate each pixel (x, V y) _x(x, y), V _y(x, y) and G (x, y) value, its computational methods are as follows:

If pixel (x, value y) be F (x, y),

V _x(x，y)＝|{F[x-1，y-1]+2F[x-1，y]+F[x-1，y+1]}-

{F[x+1，y-1]+2F[x+1，y]+F[x+1，y+1]}|

V _y(x，y)＝|{F[x-1，y-1]+2F[x，y-1]+F[x+1，y-1]}-

{F[x-1，y+1]+2F[x，y+1]+F[x+1，y+1]}|

G(x，y)＝V _x(x，y)+V _y(x，y)

Then the degrees of detail of 8 * 8 sub-pieces is

If the degrees of detail of i 8 * 8 sub-pieces is D _i(i=0,1,2,3).Determine a threshold values T ₃=1024 * (max (min (14, q/3), 6)), wherein q is a quantization parameter.Then D _iBinaryzation:

${\mathrm{DS}}_i=\left\{\begin{array}{cc}1& ,D_i>{\mathrm{<figure-callout\; id="3"\; label="T"\; filenames="S2007100506189E00021.png"\; state="\{\{state\}\}">T</figure-callout>}}_3\\ 0& ,D_i\&le;{\mathrm{<figure-callout\; id="3"\; label="T"\; filenames="S2007100506189E00021.png"\; state="\{\{state\}\}">T</figure-callout>}}_3\end{array}\right.$

The initial inter mode computation method of current macro is as follows:

(a) if 4 DS _iIn have 1 to be 1 at the most, then before the initial inter pattern of macro block be 16 * 16 patterns;

(b) if 4 DS _iIn have at least 3 to be 1, then before the initial inter pattern of macro block be 8 * 8 patterns;

(c) if 4 DS _iIn have 2 to be just at 1 o'clock:

(c1) if DS ₀=DS ₁=1 or DS ₂=DS ₃=1, then the initial inter pattern of preceding macro block is 16 * 8 patterns;

(c2) if DS ₀=DS ₂=1 or DS ₁=DS ₃=1, then the initial inter pattern of preceding macro block is 8 * 16 patterns;

(c3) if DS ₀=DS ₃=1 or DS ₁=DS ₂=1, order $V_x=\underset{y=0}{\overset{15}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{15}{\mathrm{\&Sigma;}}}{V}_x(x,y),$ $V_y=\underset{y=0}{\overset{15}{\mathrm{\&Sigma;}}}\underset{x=0}{\overset{15}{\mathrm{\&Sigma;}}}{V}_y(x,y),$ Work as V _x＞V _yThe time, then the initial inter pattern of preceding macro block is 16 * 8 patterns; Work as V _x≤ V _yThe time, then the initial inter pattern of preceding macro block is 8 * 16 patterns.

Step 3.6: the initial inter pattern of current macro is put in order pixel motion estimate.

The sub-piece of each brightness to current macro (is determined by initial inter pattern, as initial inter pattern is 16 * 8, then two sub-pieces of brightness are 16 * 8 sub-pieces) put in order the pixel motion estimation according to the mode of step 3.1, establish the estimation residual values sum SAD of the sub-piece of all brightness ₂

Step 3.7: the Intra prediction is carried out in the brightness of current macro, and calculated its originate mode.

The sub-piece of each 8 * 8 brightness in the current macro is predicted, the pixel of coboundary and left margin is all got the pixel (not getting the pixel of reconstruction frames) of present frame during prediction, the predictive mode of residual values minimum is as the predictive mode of the sub-piece of 8 * 8 brightness, and the Minimum Residual difference sum of four 8 * 8 sub-pieces of brightness is designated as SAD ₃(also claim pixel value absolute difference and, Sum of Absolute Difference is called for short SAD).

(a): if SAD ₂＞SAD ₁, and SAD ₁＞SAD ₃, then the originate mode of preceding macro block is revised as the Intra pattern;

(b): if SAD ₂＞SAD ₁, and SAD ₁≤ SAD ₃, then the originate mode of preceding macro block is revised as 16 * 16 inter patterns;

(c): if SAD ₂≤ SAD ₁, and SAD ₂＞SAD ₃, then the originate mode of preceding macro block is revised as the Intra pattern;

(d): if SAD ₂≤ SAD ₁, and SAD ₂≤ SAD ₃, then the originate mode of preceding macro block is constant;

Step 3.8: the final pattern of determining current macro.

(a) originate mode as the current macro that obtained by step 3.2 is the Intra pattern, and then the final pattern of current macro is the Intra pattern.

The final prediction mode method of its sub-piece of each 8 * 8 brightness and the sub-piece of colourity is as follows:

Calculate the predicted value of various patterns earlier by the boundary value of this sub-piece in present frame, calculate the residual values of predicted value and current sub-block then, at last the final predictive mode of the predictive mode of residual values minimum as this sub-piece.

(b) originate mode of the current macro that is obtained by step 3.3 is the SKIP pattern, and then the final pattern of current macro is the SKIP pattern.

(c) originate mode of the current macro that is obtained by step 3.4 is 16 * 16 inter patterns, and then the final pattern of current macro is 16 * 16 inter patterns, and carries out sub-pixel motion estimation, obtains final motion vector.

(d) originate mode as the current macro that obtained by step 3.7 is the inter pattern, and then the final pattern of current macro is an originate mode just, and carries out sub-pixel motion estimation, obtains final motion vector; Originate mode as the current macro that obtained by step 3.7 is the intra pattern, then the final predictive mode of the predictive mode of the prediction residual value minimum of the sub-piece of each brightness as the sub-piece of this brightness, the computational methods of the final predictive mode of the sub-piece of colourity are identical with situation (a) in this step.

Step 3.9, are all macro blocks handled in the current continuous macro block?

If all macro block modes are handled in the current continuous macro block, then proceed to step 4; Otherwise getting back to step 3.1 proceeds.

Step 4 finishes current continuous Macroblock Mode Selection process.

Present embodiment illustrates principle of the present invention with the TMS320DM64x of company of Texas Instrument (TI) series multimedia process chip to the model selection of the P-slice of AVS standard; those of ordinary skill in the art will appreciate that; other any video compression standard and any similar hardware platform are to the application of technical scheme of the present invention; although clearly do not describe here and illustrate, all should be included in protection scope of the present invention.And example described here and formula and function should be understood that to be not limited to the example and the condition of so special statement for helping reader understanding's principle of the present invention.

Claims (7)

1. the fast selecting method of a macro block mode is characterized in that, the method comprising the steps of:

Step 1, determine the size of continuous macro block during beginning according to the size of the on-chip memory of DSP, is all macroblock partitions in the frame 16 * 16 macro blocks of plurality of continuous, and all macro blocks in each continuous macro block all are in the frame with in the delegation, and distribute required sheet internal buffer;

Step 2, in present frame and the reference frame for continuous macro block carry out the needed brightness of model selection, chroma data copies to the sheet internal buffer;

Step 3 is calculated to determine the final pattern of each macro block the pattern of each macro block in the continuous macro block, comprises step:

Step 3.1 is carried out the whole pixel motion of 16 * 16 patterns and is estimated to current macro in the continuous macro block;

Step 3.2 judges according to motion estimation result whether the originate mode of current macro is the Intra pattern, if, then the originate mode of current macro as the Intra pattern, forward step 3.8 to, if not, then carry out:

Step 3.3 judges whether the originate mode of current macro is the SKIP pattern, if, then the originate mode of current macro as the SKIP pattern, forward step 3.8 to, if not, then carry out:

Step 3.4 judges whether the originate mode of current macro is 16 * 16 patterns, if, then the originate mode of current macro as 16 * 16 patterns, forward step 3.8 to, if not, then carry out:

Step 3.5, the degrees of detail of 48 * 8 sub-pieces of brightness is determined its initial interframe encoding mode in the calculating current macro;

Step 3.6 is put in order pixel motion to the initial interframe encoding mode of current macro and is estimated;

Step 3.7 is carried out the intra-frame encoding mode prediction to the brightness of current macro, and is calculated its originate mode;

Step 3.8 is determined the final pattern of current macro;

Step 4 if all macro block modes are handled in the current continuous macro block, then finishes current continuous Macroblock Mode Selection process; Otherwise getting back to step 1 proceeds.

2. the fast selecting method of macro block mode according to claim 1 is characterized in that, step 3.5 comprises step:

For each 8 * 8 sub-piece, at first calculate each pixel (x, Vx y) (x, y), Vy (x, y) and G (x, y) value, its computational methods are as follows:

If pixel (x, value y) be F (x, y),

V _x(x，y)＝|{F[x-1，y-1]+2F[x-1，y]+F[x-1，y+1]}-

{F[x+1，y-1]+2F[x+1，y]+F[x+1，y+1]}|

V _y(x，y)＝|{F[x-1，y-1]+2F[x，y-1]+F[x+1，y-1]}-

{F[x-1，y+1]+2F[x，y+1]+F[x+1，y+1]}|

G(x，y)＝V _x(x，y)+V _y(x，y)

Then the degrees of detail of 8 * 8 sub-pieces is

If the degrees of detail of i 8 * 8 sub-pieces is D _i, determine a threshold value T ₃=1024 * (max (min (14, q/3), 6)), wherein q is a quantization parameter, then D _iBinaryzation:

${\mathrm{DS}}_i=\left\{\begin{array}{c}1,D_i>T_3\\ 0,D_i\&le;T_3\end{array}\right.;$

If if (a) 4 DS _iIn have 1 to be 1 at the most, then the initial inter pattern of current macro is 16 * 16 patterns; Or,

(b) if 4 DS _iIn have at least 3 to be 1, then the initial inter pattern of current macro is 8 * 8 patterns; Or,

(c) if 4 DS _iIn have 2 to be 1 just, then the initial inter pattern of current macro is 16 * 8 patterns or 8 * 16 patterns.

3. the fast selecting method of macro block mode according to claim 2 is characterized in that, the step in the step 3.5 (c) one of may further comprise the steps:

(c1) if DS ₀=DS ₁=1 or DS ₂=DS ₃=1, then the initial inter pattern of current macro is 16 * 8 patterns; Or,

(c2) if DS ₀=DS ₂=1 or DS ₁=DS ₃=1, then the initial inter pattern of current macro is 8 * 16 patterns; Or,

(c3) if DS ₀=DS ₃=1 or DS ₁=DS ₂=1, order

Work as V _x＞V _yThe time, then the initial inter pattern of current macro is 16 * 8 patterns; Work as V _x≤ V _yThe time, then the initial inter pattern of current macro is 8 * 16 patterns.

4. the fast selecting method of macro block mode according to claim 1 is characterized in that, step 3.7 may further comprise the steps:

The sub-piece of each 8 * 8 brightness in the current macro is predicted, the pixel of coboundary and left margin is all got the pixel of present frame during prediction, the predictive mode of residual values minimum is as the predictive mode of the sub-piece of 8 * 8 brightness, and the Minimum Residual difference sum of four 8 * 8 sub-pieces of brightness is designated as SAD3, and residual values is

X wherein _{I, j}And T _{I, j}(i, the j) pixel value of position are established the estimation residual values sum SAD of the sub-piece of all brightness to be respectively in present frame and the reference frame ₂, if

(a) if SAD ₂＞SAD ₁, and SAD ₁＞SAD ₃, then the originate mode of current macro is revised as the Intra pattern;

(b) if SAD ₂＞SAD ₁, and SAD ₁≤ SAD ₃, then the originate mode of current macro is revised as 16 * 16inter pattern;

(c) if SAD ₂≤ SAD ₁, and SAD ₂＞SAD ₃, then the originate mode of current macro is revised as the Intra pattern;

(d) if SAD ₂≤ SAD ₁, and SAD ₂≤ SAD ₃, then the originate mode of current macro is constant.

5. the fast selecting method of macro block mode according to claim 1 is characterized in that, step 3.8 one of may further comprise the steps:

(a) originate mode as the current macro that obtained by step 3.2 is the Intra pattern, and then the final pattern of current macro is the Intra pattern; The final prediction mode method of its sub-piece of each 8 * 8 brightness and the sub-piece of colourity is as follows:

Calculate the predicted value of various patterns earlier by the boundary value of this sub-piece in present frame, calculate the residual values of predicted value and current sub-block then, at last the final predictive mode of the predictive mode of residual values minimum as this sub-piece; Or,

(b) originate mode of the current macro that is obtained by step 3.3 is the SKIP pattern, and then the final pattern of current macro is the SKIP pattern; Or,

(c) originate mode of the current macro that is obtained by step 3.4 is 16 * 16inter pattern, and then the final pattern of current macro is 16 * 16inter pattern, and carries out sub-pixel motion estimation, obtains final motion vector; Or,

(d) originate mode of the current macro that is obtained by step 3.7 is the inter pattern, and then the final pattern of current macro is an originate mode just, and carries out sub-pixel motion estimation, obtains final motion vector; Originate mode as the current macro that obtained by step 3.7 is the intra pattern, then the final predictive mode of the predictive mode of the prediction residual value minimum of the sub-piece of each brightness as the sub-piece of this brightness, the computational methods of the final predictive mode of the sub-piece of colourity are identical with step (a).

6. according to the fast selecting method of each described macro block mode of claim 1-5, it is characterized in that the big or small L of macro block is in the step 1:

Wherein: M ₀Be the byte-sized of effective I SRAM, the size of sheet internal buffer M is:

M=2 * (16 * L+2N) * (16+2N)+(L+1) * 16 * 26, the span of motion vector is :-N～+ N.

7. according to the fast selecting method of each described macro block mode of claim 1-5, it is characterized in that to copy to the mode of sheet internal buffer be direct memory access mode to data in the step 2.

Images