CN102752596A - Rate distortion optimization method - Google Patents

Abstract

The invention relates to a rate distortion optimization method, which comprises the following steps: A. performing rate distortion calculation by using a first calculation index to obtain a first group of distortion/code rate values; B. performing rate distortion calculation by using a second calculation index to obtain a second group of distortion/code rate values; C. acquiring a distortion/code rate value corresponding to the minimum coding cost in a first group of distortion/code rate values by adopting a first rate-distortion optimization mode; D. obtaining a plurality of candidate code rate values and corresponding coding costs according to the distortion/code rate value of the minimum coding cost; E. obtaining a second group of distortion/code rate value compensation costs corresponding to the candidate code rate values by adopting a second rate distortion optimization mode; F. and determining the minimum value of the comprehensive cost according to the coding cost and the compensation cost, and further acquiring the optimal coding mode. The rate distortion optimization method of the invention enables the rate distortion optimization result to be optimal by integrating two rate distortion optimization indexes.

Description

A kind of rate-distortion optimization method

Technical field

The present invention relates to image/video encoding and decoding field, more particularly, relate to a kind of rate-distortion optimization method.

Background technology

With international standard H264 is reference, when encoded in the basic coding unit, can select the different coding pattern.The selection of coding mode comprises the selection to infra-frame prediction mode or inter prediction mode; Also can comprise the selection of partitioning scheme to the basic coding unit (routine INTRA-4x4, INTRA-8x8, INTRA-16x16, SKIP, DIRECT, INTER-16x16, INTER-16x8, INTER-8x16, INTER-8x8, can further be divided into INTER-8x8, INTER-8x4, INTER-4x8, INTER-4x4), can also comprise selection predict blocks position (routine Intra_4x4_Vertical, Intra_4x4_Horizontal, Intra_4x4_Diagonal_Down_Left, Intra_4x4_Diagonal_Down_Right, Intra_4x4_Vertical_Right, Intra_4x4_Horizontal_Down, Intra_4x4_Vertical_Left, Intra_4x4_Horizontal_Up, Intra_4x4_DC) to INTER-8x8.Confirming of coding mode is that the percent of pass aberration optimizing realizes that wherein rate-distortion optimization is the process that following cost function J is minimized,

J(s,c,mode|QP)=D(s,c,mode|QP)+λ _modeR(s,c,mode|QP) （1）

Wherein D is a distortion value, and R is the code check value, and s and c represent former figure respectively and build the corresponding basic coding of image unit again through what encoding and decoding were handled, and mode representes the coding mode selected of basic coding unit, and QP is a quantization parameter, λ _ModeBe to be used for the Lagrangian parameter of compromise distortion value and code check value.

Rate-distortion optimization is under the fixed condition of quantization parameter QP, confirms the mode that can make above-mentioned cost function J minimum.In the H264 standard, λ _ModeBy quantization parameter QP decision, promptly have

λ _mode=0.85×2 ^(QP-12)/3 （2）

Above-mentioned rate-distortion optimization is to specify λ _ModeSituation under confirm optimum coding mode mode.But, above-mentioned λ _ModeThe statistical approximation result of definite mode during just with the distortion of MSE (mean squared error, mean square error) mensuration.And MSE can correctly express distortion, can often occur on the contrary feeling inconsistent situation with the distortion of human eye.Such as being directed against the close distorted image of some MSE values, the distortion sensation of human eye possibly have a great difference.So measure the method for distortion with MSE also perfect inadequately, therefore can cause the result of above-mentioned rate-distortion optimization is not the best yet.

So, be necessary to provide a kind of rate-distortion optimization method, to solve the existing in prior technology problem.

Summary of the invention

The technical problem that the present invention will solve is; To rate-distortion optimization method of the prior art can not be correct the expression distortion; Cause the defective of the result badly of rate-distortion optimization; Provide a kind of, make the rate-distortion optimization method that the rate-distortion optimization result is best through comprehensive two kinds of rate-distortion optimization indexs.

The technical solution adopted for the present invention to solve the technical problems is: the present invention relates to a kind of rate-distortion optimization method, it comprises step:

A, employing first parameter are carried out rate distortion calculating, obtain first group of distortion/code check value;

B, employing second parameter are carried out rate distortion calculating, obtain second group of distortion/code check value;

C, adopt the first rate-distortion optimization mode, obtain the distortion/code check value of corresponding minimum code cost in said first group of distortion/code check value;

D, according to the distortion/code check value of said minimum code cost, obtain a plurality of candidate code check values and respective coding cost;

E, adopt the second rate-distortion optimization mode, obtain the compensation cost of the second group distortion/code check value corresponding with said candidate code check value;

F, according to said coding cost and said compensation cost, confirm the minimum value of integrate-cost, and then obtain the forced coding pattern.

In rate-distortion optimization method of the present invention, said first parameter is squared difference and the index corresponding to mean square error, is specially:

$\mathrm{SSD}=\underset{(x,y)\∈A}{\mathrm{\Σ}}{(s(x,y)-c(x,y))}^2,$

Wherein SSD is squared difference and index; S representes the basic coding unit of former figure; C representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and x, y represent the location of pixels in the said basic coding unit, and A representes the pixel coverage of said basic coding unit.

In rate-distortion optimization method of the present invention, said second parameter is the structural similarity index, is specially:

SSIM(s,c)=L(s,c)·G(s.c)·H(s,c)，

$L(s,c)=\frac{2{\mathrm{\&mu;}}_s{\mathrm{\&mu;}}_c+C_1}{{\mathrm{\&mu;}}_{\mathrm{<figure-callout\; id="2"\; label="s"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">s</figure-callout>}}^2+{\mathrm{\&mu;}}_{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">c</figure-callout>}}^2+C_1},$

$G(s,c)=\frac{2{\mathrm{\&sigma;}}_s{\mathrm{\&sigma;}}_c+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="s"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">s</figure-callout>}}^2+{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">c</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">C</figure-callout>}}_2},$

$H(s,c)=\frac{{\mathrm{\&sigma;}}_{\mathrm{sc}}+C_3}{{\mathrm{\&sigma;}}_s{\mathrm{\&sigma;}}_c+C_3},$

Wherein SSIM is the structural similarity index, and s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and L representes the brightness similarity, and G representes to contrast similarity, and H representes structural similarity; μ _sBe the mean value of s interior pixel, μ _cBe the mean value of c interior pixel,

Be the variance yields of s interior pixel, Be the variance yields of c interior pixel, σ _ScBe the covariance value of respective pixel in s interior pixel and the c, C ₁, C ₂, C ₃Be constant.

In rate-distortion optimization method of the present invention, said step C is specially: obtain corresponding codes cost J in said first group of distortion/code check value through following formula _i ¹, again according to said coding cost J _i ¹Confirm the distortion/code check value of corresponding minimum code cost,

J _i ¹(s,c,mode|QP)=D _i ¹(s,c,mode|QP)+λ ₁R _i(s,c,mode|QP)，

J wherein _i ¹Be said coding cost, D _i ¹Be said first group of distortion value, R _iBe said code check value, s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and mode representes the coding mode selected of said basic coding unit, and QP is a quantization parameter, λ ₁Be to be used for the Lagrangian parameter of compromise said first group of distortion value and said code check value, i is a positive integer; Wherein

λ ₁=0.85×2 ^(QP-12)/3。

In rate-distortion optimization method of the present invention, in said step D, with the code check value of said minimum code cost, and at least two the code check values adjacent with the code check value of said minimum code cost are as said candidate code check value.

In rate-distortion optimization method of the present invention, said step e is specially: obtain the compensation cost of the second group distortion/code check value corresponding through following formula with said candidate code check value,

$J_i^2=D_i^2+k_i\&CenterDot;R_i,$

Wherein

Be said second group of distortion value, R _iBe said code check value,

Be said compensation cost, k _iBe the slope of the said curve that second group of distortion/the code check value is constituted in respective points, i is a positive integer.

In rate-distortion optimization method of the present invention, said step F is specially: obtain said integrate-cost J through following formula _i, and according to said integrate-cost J _iMinimum value confirm said forced coding pattern,

$J_i=w_1\&CenterDot;J_i^1+w_2\&CenterDot;J_i^2,$

Wherein Be said coding cost,

Be said compensation cost, w ₁, w ₂Be weight coefficient, i is a positive integer.

In rate-distortion optimization method of the present invention, when the motion vector of inter prediction was estimated, said step C was specially: obtain corresponding codes cost J in said first group of distortion/code check value through following formula _i ¹, again according to said coding cost J _i ¹Confirm the distortion/code check value of corresponding minimum code cost,

J _i ¹(s,c,mv|QP)=D _i ¹(s,c,mv|QP)+λ _motionR _i(s,c,mv|QP)，

Wherein

Be said coding cost, D _i ¹Be said first group of distortion value, R _iBe said code check value, s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and the motion vector when mv representes inter prediction, QP are quantization parameter, λ _MotionBe the Lagrangian parameter when asking motion vector, i is a positive integer; Wherein

λ _Motion=0.85 * 2 ^(QP-12)/3Or ${\mathrm{\&lambda;}}_{\mathrm{Motion}}=\sqrt{0.85\&times;2^{(\mathrm{QP}-12)/3}}.$

In rate-distortion optimization method of the present invention; Said rate-distortion optimization method also comprises step: according to the minimum value of said integrate-cost; Obtain best Lagrangian parameter; According to the Lagrangian parameter of said the best, obtain the Lagrangian parameter the when motion vector of inter prediction estimated.

In rate-distortion optimization method of the present invention, obtain the Lagrangian parameter of said the best according to following formula,

λ ^* _mode=(J-D)/R，

Wherein J is minimum integrate-cost, and D is the distortion value of said forced coding pattern, and R is the code check value of said forced coding pattern, λ ^* _ModeBe the Lagrangian parameter of said the best,

Lagrangian parameter when obtaining said motion vector and estimate to inter prediction through following formula,

λ ^* _Motion=λ ^* _ModeOr ${{\mathrm{\&lambda;}}^{*}}_{\mathrm{Motion}}=\sqrt{{{\mathrm{\&lambda;}}^{*}}_{\mathrm{Mode}}},$

λ wherein ^* _MotionLagrangian parameter when estimating for said motion vector to inter prediction.

The rate-distortion optimization method of embodiment of the present invention has following beneficial effect: through comprehensive two kinds of rate-distortion optimization indexs, make the rate-distortion optimization result best.Avoid the expression distortion that the rate-distortion optimization method of prior art can not be correct, caused the technical problem of the result badly of rate-distortion optimization.

Description of drawings

Below in conjunction with accompanying drawing and embodiment the present invention is described further, in the accompanying drawing:

Fig. 1 is the flow chart of first preferred embodiment of rate-distortion optimization method of the present invention;

Fig. 2 is the idiographic flow block diagram of first preferred embodiment of rate-distortion optimization method of the present invention;

Fig. 3 is for obtaining the sketch map of candidate's code check value in the rate-distortion optimization method of the present invention.

Embodiment

In order to make the object of the invention, technical scheme and advantage clearer,, the present invention is further elaborated below in conjunction with accompanying drawing and embodiment.Should be appreciated that specific embodiment described herein only in order to explanation the present invention, and be not used in qualification the present invention.

Please with reference to Fig. 1, Fig. 1 is the flow chart of first preferred embodiment of rate-distortion optimization method of the present invention.This rate-distortion optimization method starts from:

Step 101 adopts first parameter to carry out rate distortion and calculates, and obtains first group of distortion/code check value;

Step 102 adopts second parameter to carry out rate distortion and calculates, and obtains second group of distortion/code check value;

Step 103 adopts the first rate-distortion optimization mode, obtains the distortion/code check value of corresponding minimum code cost in first group of distortion/code check value,

Step 104, the distortion/code check value according to the minimum code cost obtains a plurality of candidate code check values and respective coding cost;

Step 105 adopts the second rate-distortion optimization mode, obtains the compensation cost of the second group distortion/code check value corresponding with candidate code check value;

Step 106 according to said coding cost and said compensation cost, is confirmed the minimum value of integrate-cost, and then is obtained the forced coding pattern.

Please with reference to Fig. 2, Fig. 2 is the idiographic flow block diagram of first preferred embodiment of rate-distortion optimization method of the present invention.Pass through the practical implementation process of the detailed explanation first rate-distortion optimization method of the present invention of Fig. 2 below.

At first; Before step 101, the input coding elementary cell is selected all coding modes of this coding elementary cell to code device then one by one; To each coding mode; Accomplish respective coding and decoding processing, the elementary cell that obtains encoding is corresponding builds image again, and the processing method of Code And Decode can be carried out according to various video encoding standards.Such as, encoding process can be carried out following each step successively: confirm predict blocks, residual computations, dct transform (discrete cosine transform, Discrete Cosine Transform), quantification, entropy coding.Decoding processing then is the reverse operating to above-mentioned each step.The concrete grammar that rate-distortion optimization method of the present invention is handled this Code And Decode does not limit.

Come step 101 subsequently, in step 101, adopt first parameter to build the corresponding basic coding of image unit again, carry out rate distortion and calculate, obtain first group of distortion/code check value to the basic coding unit of former figure with through what encoding and decoding were handled.In the present embodiment, first parameter is SSD (squared difference with, the sum of squared difference) index of corresponding MSE (mean square error, mean squared error).This rate distortion calculates and is specially:

$\mathrm{SSD}=\underset{(x,y)\&Element;A}{\mathrm{\&Sigma;}}{(s(x,y)-c(x,y))}^2,$

Wherein SSD is squared difference and index; S representes the basic coding unit of former figure; C representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and x, y represent each locations of pixels in the said basic coding unit, and A representes the pixel coverage of basic coding unit.

Through the calculating of following formula, s and the c corresponding according to each coding mode obtain corresponding code check value R _iWith the distortion value D that expresses with the SSD index _i ¹, can obtain first group of distortion/code check value like this, be designated as (D _i ¹, R _i), wherein i is a positive integer, the maximum of i is to the number of coding mode that should the basic coding unit.

Come step 102 subsequently, in step 102, adopt second parameter to build the corresponding basic coding of image unit again, carry out rate distortion and calculate, obtain second group of distortion/code check value to the basic coding unit of former figure with through what encoding and decoding were handled.In the present embodiment, second parameter is SSIM (structural similarity, a structural similarity) index.This rate distortion calculates and is specially:

SSIM(s,c)=L(s,c)·G(s.c)·H(s,c)，

$L(s,c)=\frac{2{\mathrm{\&mu;}}_s{\mathrm{\&mu;}}_c+C_1}{{\mathrm{\&mu;}}_s^2+{\mathrm{\&mu;}}_{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">c</figure-callout>}}^2+C_1},$

$G(s,c)=\frac{2{\mathrm{\&sigma;}}_s{\mathrm{\&sigma;}}_c+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="s"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">s</figure-callout>}}^2+{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">c</figure-callout>}}^2+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">C</figure-callout>}}_2},$

$H(s,c)=\frac{{\mathrm{\&sigma;}}_{\mathrm{sc}}+C_3}{{\mathrm{\&sigma;}}_s{\mathrm{\&sigma;}}_c+C_3},$

Wherein SSIM is the structural similarity index, and s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and L representes the brightness similarity, and G representes to contrast similarity, and H representes structural similarity; μ _sBe the mean value of s interior pixel, μ _cBe the mean value of c interior pixel,

Be the variance yields of s interior pixel,

Be the variance yields of c interior pixel, σ _ScBe the covariance value of respective pixel in s interior pixel and the c, C ₁, C ₂, C ₃Be constant, C ₁, C ₂, C ₃Value can obtain through experiment.

Through the calculating of following formula, s and the c corresponding according to each coding mode can obtain corresponding code check value R _iWith the distortion value of expressing with the SSIM index

Can obtain second group of distortion/code check value like this, be designated as

Wherein i is a positive integer, and the maximum of i is to the number of coding mode that should the basic coding unit.

Because adopt calculating of SSD index and SSIM index to calculate the just different calculation methods of distortion value, and to code check value R _iSo did not influence is first group of distortion/code check value (D _i ¹, R _i) and second group of distortion/code check value

In code check value R _iIdentical.

Come step 103 subsequently, in step 103, adopt the first rate-distortion optimization mode, obtain the distortion/code check value of corresponding minimum code cost in first group of distortion/code check value.Be specially, obtain corresponding codes cost J in first group of distortion/code check value through following formula _i ¹, again according to coding cost J _i ¹Confirm the distortion/code check value of corresponding minimum code cost,

J _i ¹(s,c,mode|QP)=D _i ¹(s,c,mode|QP)+λ ₁R _i(s,c,mode|QP)，

J wherein _i ¹Be coding cost, D _i ¹Be first group of distortion value, R _iBe the code check value, s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and mode representes the coding mode selected of said basic coding unit, and QP is a quantization parameter, λ ₁Be to be used for the Lagrangian parameter (i.e. the first mode parameter) of compromise said first group of distortion value and said code check value, i is a positive integer; Wherein

λ ₁=0.85×2 ^(QP-12)/3。

Come step 104 subsequently, in step 104, the distortion/code check value according to the minimum code cost obtains a plurality of candidate code check values and respective coding cost.Specifically as shown in Figure 3, Fig. 3 is for obtaining the sketch map of candidate's code check value in the rate-distortion optimization method of the present invention.Abscissa is code check value R among Fig. 3 _i, ordinate is distortion value D _i, comprising first group of distortion/code check value (D _i ¹, R _i) matched curve and second group of distortion/code check value

Matched curve.According to first group of distortion/code check value (D _i ¹, R _i) the code check value of the minimum code cost that obtains, and at least two the code check values adjacent with the code check value of this minimum code cost are as candidate code check value.As among Fig. 3

The coding cost Minimum is then with code check value R ₃, and with code check value R ₃Adjacent code check value R ₂With code check value R ₄As candidate's code check value, candidate's code check value R ₂The corresponding codes cost is J ₂ ¹, candidate's code check value R ₃The corresponding codes cost is J ₃ ¹, candidate's code check value R ₄The corresponding codes cost is J ₄ ¹Certainly also can select more candidate's code check value here, as will with code check value R ₃4 adjacent code check value R ₁, code check value R ₂, code check value R ₄And code check value R ₅All as candidate's code check value.On the other hand, if R ₃The left side or the right during not corresponding to the adjacent code check value of coding mode, then only select R ₃With corresponding monolateral adjacent code check value as candidate's code check value.

Come step 105 subsequently, in step 105, adopt the second rate-distortion optimization mode, obtain the compensation cost of the second group distortion/code check value corresponding with candidate code check value.

With the data point among Fig. 3 is example, and the code check value sorts from small to large and is followed successively by R ₁, R ₂, R ₃, R ₄And R ₅, such curve that second group of distortion/the code check value is constituted is at candidate's code check value R ₂Corresponding point (D ₂ ², R ₂) slope be:

${\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">k</figure-callout>}}_2=-\frac{D_3^2-D_1^2}{R_3-R_1};$

Second group of curve that distortion/the code check value is constituted is at candidate's code check value R ₃Pairing point (D ₃ ², R ₃) slope be:

$k_3=-\frac{D_4^2-{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">D</figure-callout>}}_2^2}{R_4-{\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">R</figure-callout>}}_2};$

Second group of curve that distortion/the code check value is constituted is at candidate's code check value R ₄Pairing point (D ₄ ², R ₄) slope be:

$k_4=-\frac{D_5^2-D_3^2}{R_5-R_3};$

Obtain the compensation cost of the second group distortion/code check value corresponding again with candidate code check value through following formula,

${\mathrm{<figure-callout\; id="2"\; label="J\; i"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">J</figure-callout>}}_i^{}={\mathrm{<figure-callout\; id="2"\; label="D\; i"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">D</figure-callout>}}_i^{}+k_i\&CenterDot;R_i,$

Wherein Be second group of distortion value, R _iBe the code check value,

Be compensation cost, k _iFor the said curve that second group of distortion/the code check value is constituted the slope of respective points (above try to achieve), i is a positive integer.Can calculate candidate's code check value R so respectively ₂The compensation cost

Candidate's code check value R ₃The compensation cost

And candidate's code check value R ₄The compensation cost In this step, second group of curve that distortion/the code check value is constituted obtained in also available other modes of the slope of respective points, as asking for the matched curve of R～D earlier, utilizes this matched curve slope calculations again.Therefore second group of curve that distortion/the code check value is constituted do not limit protection scope of the present invention at the concrete obtain manner of the slope of respective points.

Come step 106 subsequently, according to coding cost J _i ¹And compensation cost

Confirm integrate-cost J _iMinimum value, and then obtain the forced coding pattern.

Obtain integrate-cost J through following formula _i:

$J_i=w_1\&CenterDot;J_i^1+{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">w</figure-callout>}}_2\&CenterDot;{\mathrm{<figure-callout\; id="2"\; label="J\; i"\; filenames="HDA00001854334000022.png"\; state="\{\{state\}\}">J</figure-callout>}}_i^{},$

Wherein Be the coding cost,

Be compensation cost, w ₁, w ₂Be weight coefficient, can get fixed value or set that i is a positive integer by the user.At last according to integrate-cost J _iMinimum value confirm forced coding pattern output.

As second preferred embodiment of rate-distortion optimization method of the present invention, the rate distortion when rate-distortion optimization method of the present invention also can be estimated the motion vector of inter prediction is optimized.Be with the difference of first preferred embodiment,

Obtain corresponding codes cost J in first group of distortion/code check value through following formula _i ¹, again according to coding cost J _i ¹Confirm the distortion/code check value of corresponding minimum code cost,

J _i ¹(s,c,mv|PQ)=D _i ¹(s,c,mv|PQ)+λ _motionR _i(s,c,mv|PQ)，

Wherein

Be coding cost, D _i ¹Be first group of distortion value, R _iBe the code check value, s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and the motion vector when mv representes inter prediction, QP are quantization parameter, λ _MotionBe the Lagrangian parameter (the first mode parameter) when asking motion vector, i is a positive integer; Wherein

λ _Motion=0.85 * 2 ^(QP-12)/3Or ${\mathrm{\&lambda;}}_{\mathrm{Motion}}=\sqrt{0.85\&times;2^{(\mathrm{QP}-12/3)}}.$

Other steps and first preferred embodiment are same or similar, specifically see also first preferred embodiment of the present invention.

The 3rd preferred embodiment as rate-distortion optimization method of the present invention; Rate-distortion optimization method of the present invention also can be according to the minimum value of the integrate-cost in first preferred embodiment; Obtain the Lagrangian parameter the when motion vector of inter prediction estimated, the rate distortion so that the motion vector of inter prediction is estimated carries out better optimize.

At first, obtain best Lagrangian parameter, be specially according to the minimum value of the integrate-cost in first preferred embodiment:

λ ^* _mode=(J-D)/R，

Wherein J is minimum integrate-cost, and D is the distortion value of forced coding pattern, and R is the code check value of forced coding pattern, λ ^* _ModeBe the Lagrangian parameter of the best.

Then according to the Lagrangian parameter of the best, obtain the Lagrangian parameter the when motion vector of inter prediction estimated, be specially

λ ^* _Motion=λ ^* _ModeOr ${{\mathrm{\&lambda;}}^{*}}_{\mathrm{Motion}}=\sqrt{{{\mathrm{\&lambda;}}^{*}}_{\mathrm{Mode}}},$

λ wherein ^* _MotionLagrangian parameter when the motion vector of inter prediction is estimated.

In sum, rate-distortion optimization method of the present invention makes the rate-distortion optimization result best through comprehensive two kinds of rate-distortion optimization indexs.Used the combination of SSD index and SSIM index in an embodiment of the present invention; Can certainly carry out complex optimum with other index; For example: SSD can use SAD (absolute difference sum; Sum of Absolute Differences) or SATD replacements such as (absolute value summation again after the Hadamard conversion, Sum of Absolute Transformed Differences), SSIM can use MS-SSIM (multiple dimensioned SSIM; Multi-Scale SSIM) or VIF replacements such as (visual information fidelity, Visual Information Fidelity); Rate-distortion optimization method of the present invention can be looked for the forced coding pattern from all coding modes of coding elementary cell simultaneously, also can from a part of coding mode of appointment, look for the forced coding pattern; Rate-distortion optimization can be realized according to the type (being I frame, P frame or B frame) of frame coding is independent separately in addition; Rate-distortion optimization method of the present invention has been avoided the expression distortion that the rate-distortion optimization method of prior art can not be correct well, causes the technical problem of the result badly of rate-distortion optimization.

The above is merely embodiments of the invention; Be not so limit claim of the present invention; Every equivalent structure transformation that utilizes specification of the present invention and accompanying drawing content to be done, or directly or indirectly be used in other relevant technical fields, all in like manner be included in the scope of patent protection of the present invention.

Claims (10)

1. a rate-distortion optimization method is characterized in that, comprises step:

A, employing first parameter are carried out rate distortion calculating, obtain first group of distortion/code check value;

B, employing second parameter are carried out rate distortion calculating, obtain second group of distortion/code check value;

C, adopt the first rate-distortion optimization mode, obtain the distortion/code check value of corresponding minimum code cost in said first group of distortion/code check value;

D, according to the distortion/code check value of said minimum code cost, obtain a plurality of candidate code check values and respective coding cost;

E, adopt the second rate-distortion optimization mode, obtain the compensation cost of the second group distortion/code check value corresponding with said candidate code check value;

F, according to said coding cost and said compensation cost, confirm the minimum value of integrate-cost, and then obtain the forced coding pattern.

2. rate-distortion optimization method according to claim 1 is characterized in that, said first parameter is squared difference and the index corresponding to mean square error, is specially:

$\mathrm{SSD}=\underset{(x,y)\&Element;A}{\mathrm{\&Sigma;}}{(s(x,y)-c(x,y))}^2,$

Wherein SSD is squared difference and index; S representes the basic coding unit of former figure; C representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and x, y represent the location of pixels in the said basic coding unit, and A representes the pixel coverage of said basic coding unit.

3. rate-distortion optimization method according to claim 1 is characterized in that, said second parameter is the structural similarity index, is specially:

SSIM(s,c)=L(s,c)·G(s.c)·H(s,c)，

$L(s,c)=\frac{2{\mathrm{\&mu;}}_s{\mathrm{\&mu;}}_c+C_1}{{\mathrm{\&mu;}}_s^2+{\mathrm{\&mu;}}_c^2+C_1},$

$G(s,c)=\frac{2{\mathrm{\&sigma;}}_s{\mathrm{\&sigma;}}_c+C_2}{{\mathrm{\&sigma;}}_s^2+{\mathrm{\&sigma;}}_c^2+C_2},$

$H(s,c)=\frac{{\mathrm{\&sigma;}}_{\mathrm{sc}}+C_3}{{\mathrm{\&sigma;}}_s{\mathrm{\&sigma;}}_c+C_3},$

Wherein SSIM is the structural similarity index, and s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and L representes the brightness similarity, and G representes to contrast similarity, and H representes structural similarity; μ _sBe the mean value of s interior pixel, μ _cBe the mean value of c interior pixel,

Be the variance yields of s interior pixel,

Be the variance yields of c interior pixel, σ _ScBe the covariance value of respective pixel in s interior pixel and the c, C ₁, C ₂, C ₃Be constant.

4. rate-distortion optimization method according to claim 1 is characterized in that, said step C is specially: obtain corresponding codes cost J in said first group of distortion/code check value through following formula _i ¹, again according to said coding cost J _i ¹Confirm the distortion/code check value of corresponding minimum code cost,

J _i ¹(s,c,mode|QP)=D _i ¹(s,c,mode|QP)+λ ₁R _i(s,c,mode|QP)，

J wherein _i ¹Be said coding cost, D _i ¹Be said first group of distortion value, R _iBe said code check value, s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and mode representes the coding mode selected of said basic coding unit, and QP is a quantization parameter, λ ₁Be to be used for the Lagrangian parameter of compromise said first group of distortion value and said code check value, i is a positive integer; Wherein

λ ₁=0.85×2 ^(QP-12)/3。

5. rate-distortion optimization method according to claim 1 is characterized in that, in said step D, with the code check value of said minimum code cost, and at least two the code check values adjacent with the code check value of said minimum code cost are as said candidate code check value.

6. rate-distortion optimization method according to claim 1 is characterized in that, said step e is specially: obtain the compensation cost of the second group distortion/code check value corresponding through following formula with said candidate code check value,

$J_i^2=D_i^2+k_i{\&CenterDot;R}_i,$

Wherein

Be said second group of distortion value, R _iBe said code check value,

Be said compensation cost, k _iBe the slope of the said curve that second group of distortion/the code check value is constituted in respective points, i is a positive integer.

7. rate-distortion optimization method according to claim 1 is characterized in that, said step F is specially: obtain said integrate-cost J through following formula _i, and according to said integrate-cost J _iMinimum value confirm said forced coding pattern,

$J_i=w_1\&CenterDot;J_i^1+w_2\&CenterDot;J_i^2,$

Wherein

Be said coding cost, Be said compensation cost, w ₁, w ₂Be weight coefficient, i is a positive integer.

8. rate-distortion optimization method according to claim 1 is characterized in that, when the motion vector of inter prediction was estimated, said step C was specially: obtain corresponding codes cost J in said first group of distortion/code check value through following formula _i ¹, again according to said coding cost J _i ¹Confirm the distortion/code check value of corresponding minimum code cost,

J _i ¹(s,c,mv|QP)=D _i ¹(s,c,mv|QP)+λ _motionR _i(s,c,mv|QP)，

Wherein Be said coding cost, D _i ¹Be said first group of distortion value, R _iBe said code check value, s representes the basic coding unit of former figure, and c representes to build the corresponding basic coding of image unit again through what encoding and decoding were handled, and the motion vector when mv representes inter prediction, QP are quantization parameter, λ _MotionBe the Lagrangian parameter when asking motion vector, i is a positive integer; Wherein

λ _Motion=0.85 * 2 ^(QP-12)/3Or ${\mathrm{\&lambda;}}_{\mathrm{Motion}}=\sqrt{0.85\&times;2^{(\mathrm{QP}-12/3)}}.$

9. rate-distortion optimization method according to claim 1; It is characterized in that; Said rate-distortion optimization method also comprises step: according to the minimum value of said integrate-cost; Obtain best Lagrangian parameter,, obtain the Lagrangian parameter the when motion vector of inter prediction estimated according to the Lagrangian parameter of said the best.

10. rate-distortion optimization method according to claim 9 is characterized in that, obtains the Lagrangian parameter of said the best according to following formula,

λ ^* _mode=(J-D)/R，

Wherein J is minimum integrate-cost, and D is the distortion value of said forced coding pattern, and R is the code check value of said forced coding pattern, λ ^* _ModeBe the Lagrangian parameter of said the best,

Lagrangian parameter when obtaining said motion vector and estimate to inter prediction through following formula,

λ ^* _Motion=λ ^* _ModeOr ${{\mathrm{\&lambda;}}^{*}}_{\mathrm{Motion}}=\sqrt{{{\mathrm{\&lambda;}}^{*}}_{\mathrm{Mode}}},$

λ wherein ^* _MotionLagrangian parameter when estimating for said motion vector to inter prediction.

Images