CN104219526A - HEVC rate distortion optimization algorithm based on just-noticeable perception quality judging criterion - Google Patents

Abstract

The invention relates to an HEVC rate distortion optimization algorithm based on the just-noticeable perception quality judging criterion. The algorithm is characterized by including analyzing the motion models and static texture features of each macroblock of each frame, acquiring the perception quality type of the current macroblock, and acquiring the image salient region; calculating a just-noticeable distortion threshold on the basis of a visual salient region; calculating the perception quality on the basis of a just-noticeable distortion model; performing rate distortion optimization on the basis of the perception quality on the basis of the just-noticeable distortion model. The algorithm is reasonable in design, the HEVC rate distortion optimization is performed by the just-noticeable perception quality judging criterion, defects that MSE (mean-square errors) are uses for measuring the video distortion estimation standard can be overcome, and the final encoding effect can meet the visual subjective perception quality better; meanwhile, more noises can be tolerated on the premise that the subjective quality is maintained, unnecessary perception redundancy is removed, the compressing efficiency is improved, and the code rate of encoded files is decreased.

Description

Based on the HEVC rate-distortion optimization algorithm just can examining perceived quality decision rule

Technical field

The invention belongs to technical field of video coding, especially a kind of HEVC rate-distortion optimization algorithm based on perceived quality decision rule just can be examined.

Background technology

In recent years, along with the fast development of national economy, the progress of technology and people's improving constantly video quality demands, high definition/ultra high-definition video coding technique becomes as the basic core technology of the business such as future home movie theatre, digital broadcast television, Internet video, high-definition movie the focus that industry pays close attention to.For high definition/ultra high-definition video communication, existing video encoding standard compares distance certain in addition at compression ratio with actual application demand.For this reason, International Organization for standardization ISO/IEC (MPEG) and ITU-T starts the planning of generation digital video compression standard---Video coding (High Efficiency Video Coding efficiently, HEVC), target is on H.264/AVC high-grade basis, and compression efficiency is enhanced about more than once.

Efficient Video coding (HEVC) always has two links in loop filtering process: block-eliminating effect filtering and self adaptation sampling point compensate SAO.Wherein, self adaptation sampling point compensation SAO can be divided into banded compensation (Band Offset, BO) and the large class of edge compensation (Edge offset, EO) two further.Edge compensation algorithm (EO) compensates mainly for the profile of object each in image, needs from level, vertical, left-leaning unity slope and selects four class adjacent encoder blocks of Right deviation unity slope a kind ofly to carry out comparing of the value of current pixel point and the value of adjacent two pixels.Banded backoff algorithm (BO) is mainly used in compensating the color of object inside each in image and lines information, its division compensating type is completely based on the amplitude of pixel itself, that is image pixel intensities is divided into 32 grades from 0 to maximum by HEVC, the selection of percent of pass aberration optimizing, wherein the pixel compensation of 4 successives finally will write code stream.

HEVC encoder, according to picture material, adopts the method for rate-distortion optimization, chooses best coding mode in alternative modes numerous with interframe in frame.Although to a certain extent, rate distortion mode adjudging can make cataloged procedure become complicated, just because of the application of rate-distortion optimization technology, encoder can obtain optimum prediction information as far as possible, thus ensure that picture quality, improves the overall performance of encoder.

The rate distortion framework of conventional video coding comprises HEVC and all uses mean square error MSE to calculate as distortion value.Although in most cases MSE can reflect the real quality of image, still have certain situation to be that MSE does not reflect, such as salt-pepper noise can produce huge interference to picture, and the MSE of whole two field picture may not be very large.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, provide a kind of reasonable in design, subjective visual quality is high and can remove more perception redundancies based on the HEVC rate-distortion optimization algorithm just can examining perceived quality decision rule.

The present invention solves existing technical problem and takes following technical scheme to realize:

Based on the HEVC rate-distortion optimization algorithm just can examining perceived quality decision rule, comprise the following steps:

Step 1, before the coding side of efficient video codec carries out mode adjudging, analyze the motor pattern of each macro block in each frame and static textural characteristics, obtain the perceived quality type of current macro, and obtain salient region of image according to different motion states;

Step 2, to calculate view-based access control model salient region according to specific image region just can examine distortion threshold;

Step 3, according to view-based access control model salient region just can examine distortion threshold calculate based on the perceived quality just can examining distortion model;

Step 4, basis carry out rate-distortion optimization based on the perceived quality just can examining distortion model.

And described step 1 perceived quality type adopts following Mathematical Modeling to obtain:

${\mathrm{mp}}_k^t=\left\{\begin{array}{ccc}\mathrm{Normal\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{0,k}^t,s_k^t\≠\mathrm{SS}\\ \mathrm{Aliased\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{1,k}^t,s_k^t\≠\mathrm{SS}\\ \mathrm{Hysteresis\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}\≠p_{2,k}^t,s_k^t=\mathrm{SS}\\ \mathrm{Background}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{2,k}^t,s_k^t\≠\mathrm{SS}\end{array}\right.$

In formula, for the perceived quality type of current macro, Normal Pattern is normal perceived quality type, and Aliased Pattern is distortion-aware quality type, and Hysteresis Pattern is delayed perceived quality type, and Background is static perceived quality type, with be respectively the probability of normal condition, distortion status, inactive state appearance, for with vector form, for status attribute descriptor, SS represents inactive state.

And described salient region of image is the region that the macro block of Normal Pattern type and the macro block of Hysteresis Pattern type combine formation.

And the distortion threshold of just can examining of described step 2 view-based access control model salient region calculates according to the following equation:

In above-mentioned formula, FJND be view-based access control model salient region just can examine distortion threshold, T _basic(k, n, i, j), F _lum, F _contrast, F _temporaland F _foveabasic threshold value, intensity modifier value, contrast correction value, time-domain correction value and marking area correction value respectively.

And described step 3 is obtained by PSNR-HA and PSNR-HMA two kinds of distortion criterion weightings based on the perceived quality just can examining distortion model, and its computational methods are as follows:

(1) for given reference block A and distortion block B, the difference of both calculating with the mean value of reference block A and distortion block B coefficient respectively;

(2) correction matrix C=B+Delt is obtained;

(3) correction factor is calculated $\mathrm{\ρ}=\frac{\mathrm{\Σ}(A-\stackrel{\‾}{A})(C-\stackrel{\‾}{C})}{\mathrm{\Σ}{(C-\stackrel{\‾}{C})}^2};$

(4) revised macro block D=C × ρ is calculated;

(5) calculating just can examine the revised distortion value MSE of distortion model _hVS, computational methods are as follows:

Wherein coeff _o(i, j) and coeff _d(i, j) represents the pixel value of reference block and reconstructed block correspondence position respectively, and what jnd (i, j) then represented second correspondence position calculated just can examine distortion threshold;

(6) if M ₁> M ₂then

$M_1=M_2+\left\{\begin{array}{c}(M_1-M_2)\mathrm{coef}1,\mathrm{\&rho;}<1\\ (M_1-M_2)\mathrm{coef}2,\mathrm{\&rho;}\&GreaterEqual;1\end{array}\right.;$ Finally obtain

MSE _HVS＝M ₁+Delt ²×coef3

In above formula, coef1, coef2, coef3 represent the modifying factor to perceptual error, get 0.002,0.25 and 0.04 respectively according to experiment experience;

(7) PSNR-HA is obtained according to the above-mentioned perceived quality distortion calculated:

$\mathrm{PSNR}-\mathrm{HA}=10{\log }_{10}\left(\frac{{255}^2}{M}\right)$

For coloured image, M is Y, Cr, Cb luminance component, the average distortion that the perceptual distortion weighting of two color components obtains, as shown in the formula calculating:

M＝(M _Y+M _Cb×coef4+M _Cr×coef4)/(1+2×coef4)

M _y, M _cb, M _cra luminance component Y respectively, the perceptual distortion of two color components Cb, Cr, weight coefficient coef4 is 0.5;

The correction of PSNR-HMA is at calculating MSE _hVStime do corresponding correction with distortion model just can be examined.

And the method that described step 4 carries out rate-distortion optimization is: just will can examine perceived quality based on salient region of image and melt and be combined on R-λ model, and obtain as drag:

$\frac{\mathrm{dJ}}{\mathrm{dR}}=\frac{\mathrm{dQ}}{\mathrm{dR}}-\mathrm{\&lambda;}=0$

In above formula, J is cost function, then λ=dQ/dR, for each macro block, and calculation cost function:

j ₁＝q ₁(QP ₁)-λ·r ₁(QP ₁)for?coding?block#1

j ₂＝q ₂(QP ₂)-λ·r ₂(QP ₂)for?coding?block#2

……

j _N＝q _N(QP _N)-λ·r _N(QP _N)for?coding?block#N

$\max \left\{J\right\}=\max ({\mathrm{\&Sigma;}}_{n=1}^N{q}_n\left({\mathrm{QP}}_n\right)-\mathrm{\&lambda;}\&CenterDot;{\mathrm{\&Sigma;}}_{n=1}^N{r}_i\left({\mathrm{QP}}_n\right))$

In above formula, QP is quantization step, compares by search the coding mode obtaining making max{J} minimum.

Advantage of the present invention and good effect are:

The present invention is reasonable in design, it adopts and carries out HEVC rate-distortion optimization based on just examining perceived quality decision rule, mean square error MSE can be overcome as the deficiency weighing video distortion evaluation criterion, final encoding efficiency is made more to meet the subjective perceptual quality of human eye, meanwhile, more noise can be tolerated under the prerequisite that the video after coding does not reduce at subjective quality, remove unnecessary perception redundancy, thus improve compression efficiency, reduce the code check of the rear file of coding.

Accompanying drawing explanation

Fig. 1 is general frame figure of the present invention;

Fig. 2 is the video interception that embodiment provides;

Fig. 3 is the remarkable figure to obtaining after Fig. 2 process.

Embodiment

Below in conjunction with accompanying drawing, the embodiment of the present invention is further described.

Based on the HEVC rate-distortion optimization algorithm just can examining perceived quality decision rule, as shown in Figure 1, comprise the following steps:

Step 1, before the coding side of efficient video codec carries out mode adjudging, analyze the motor pattern of each macro block in each frame and static textural characteristics, draw the perceived quality type of current macro, and obtain salient region of image according to different motion states.

In this step, for each coding unit CU of each frame, and its each divided block, first set up the status attribute descriptor of complete set (representing the state of the kth macro block in a frame of t), as follows:

$s_k^t=\left\{\begin{array}{ccc}\mathrm{Normal\; state}\left(\mathrm{NS}\right)& \mathrm{if}& {\mathrm{MV}}_k^t\&NotEqual;0,R_k^t\&le;1\\ \mathrm{Aliased\; state}\left(\mathrm{AS}\right)& \mathrm{if}& {\mathrm{MV}}_k^t\&NotEqual;0,R_k^t>1\\ \mathrm{Stationary\; state}& \mathrm{if}& {\mathrm{MV}}_k^t=0\end{array}\right.$ Wherein

$R_k^t=\frac{\mathrm{Temporal\; residual\; energy}}{\mathrm{Texture\; energy}}=\frac{{\mathrm{\&Sigma;}}_{q\&Element;B_k^t}{\{B_k^t\left(q\right)-B_k^t(q+{\mathrm{MV}}_k^t)\}}^2}{{\mathrm{\&Sigma;}}_{q\&Element;B_k^t}{\left\{B_k^t\left(q\right)\right\}}^2-\frac{1}{N\&times;N}{\left\{{\mathrm{\&Sigma;}}_{q\&Element;B_k^t}\left(B_k^t\left(q\right)\right)\right\}}^2}$

In formula, NS, AS, SS represent normal condition, distortion status, inactive state respectively, represent motion vector, the ratio representing current block residual energy and texture energy ratio, represent the kth macro block in a frame of t, the size of q to be corresponding quantization parameter N × N be macro block.

With vector form represent status attribute descriptor as follows:

$I_k^t=\left\{\begin{array}{cccc}{\left[\mathrm{0,1,1}\right]}^T& \mathrm{if}& s_k^t=\mathrm{NS}& \\ {[1,\mathrm{0,1}]}^T& \mathrm{if}& s_k^t=\mathrm{AS}& L_k^t=\underset{r=t-T}{\overset{t}{\mathrm{\&Sigma;}}}{I}_k^t={[L_{0,k}^t,L_{1,k}^t,L_{2,k}^t]}^T\\ {\left[\mathrm{1,1,0}\right]}^T& \mathrm{if}& s_k^t=\mathrm{SS}& \end{array}\right.$

Wherein three values corresponding three kinds of states situation about not occurring respectively, namely 0 represents and does not occur, and 1 represents and occurs.

At a time, represent the weight of the probability that NS, AS and SS occur respectively, three uses vector form represent; with represent the probability that in three, state occurs, be expressed as vector form as follows:

$P_k^t\&equiv;\left[\begin{array}{c}{p}_{0,k}^t\\ p_{1,k}^t\\ p_{2,k}^t\end{array}\right]=\frac{1}{\left|W_k^t\right|}\left[\begin{array}{c}{\mathrm{\&omega;}}_{0,k}^t\\ {\mathrm{\&omega;}}_{1,k}^t\\ {\mathrm{\&omega;}}_{2,k}^t\end{array}\right],$ Wherein $W_k^t=\left[\begin{array}{c}{\mathrm{\&omega;}}_{0,k}^t\\ {\mathrm{\&omega;}}_{1,k}^t\\ {\mathrm{\&omega;}}_{2,k}^t\end{array}\right]=\left[\begin{array}{c}{\mathrm{\&omega;}}_{0,k}^t\&CenterDot;{\exp }^{-{\mathrm{\&eta;L}}_k^t\left(0\right)}\\ {\mathrm{\&omega;}}_{1,k}^t\&CenterDot;{\exp }^{-{\mathrm{\&eta;L}}_k^t\left(1\right)}\\ {\mathrm{\&omega;}}_{2,k}^t\&CenterDot;{\exp }^{-{\mathrm{\&eta;L}}_k^t\left(2\right)}\end{array}\right]$

η is a constant being greater than 1, represents iterative rate;

Draw each macro block probability vector according to above-mentioned parameter, judge the perceived quality type belonging to it according to following criterion as follows:

${\mathrm{mp}}_k^t=\left\{\begin{array}{ccc}\mathrm{Normal\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{0,k}^t,s_k^t\&NotEqual;\mathrm{SS}\\ \mathrm{Aliased\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{1,k}^t,s_k^t\&NotEqual;\mathrm{SS}\\ \mathrm{Hysteresis\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}\&NotEqual;p_{2,k}^t,s_k^t=\mathrm{SS}\\ \mathrm{Background}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{2,k}^t,s_k^t\&NotEqual;\mathrm{SS}\end{array}\right.$

Background in formula represents that picture still is motionless, is background parts;

Think in this method and only belong to Normal Pattern and Hysteresis Pattern two kinds of situations both attracting attentivenesss, the details differentiating picture can be divided again.The region that the macro block belonging to these two kinds of patterns is combined and salient region.Fig. 2 gives a video interception, and Fig. 3 is the remarkable figure obtained by this step.

Step 2, to calculate view-based access control model salient region according to specific image region just can examine distortion threshold.

This method can attract the attentiveness of beholder for moving region in video content, and the visual sensitivity of human eye outwards declines gradually along with vision central fovea, saliency region and tradition combine so just can be examined distortion model, further excavation video-aware redundancy, improves compression efficiency under the prerequisite ensureing viewing quality.

The distortion threshold FJND that just can examine of view-based access control model salient region calculates according to the following equation:

In above-mentioned formula, T _basic(k, n, i, j), F _lum, F _contrast, F _temporaland F _foveabe basic threshold value, intensity modifier value, contrast correction value, time-domain correction value and marking area correction value respectively, wherein former three is for the threshold value of still image and modifying factor, after two modifying factors for video.The calculating of each factor is respectively:

(1) basic threshold value T _basic(k, n, i, j):

In formula, (n, i, j) represents (i, the j) of n-th piece individual position, and s illustrates that space adds up the parameter of effect, gets 0.25 by empirical value; A, b, c, r are that constant equals 1.33,0.11,0.18 and 0.6 respectively.φ _iand φ _jit is the normalization factor of dct transform ${\mathrm{\&phi;}}_m=\left\{\begin{array}{c}\sqrt{1/N},m=0\\ \sqrt{2/N},m>0\end{array}\right.,$ ω _ijit is frequency ${\mathrm{\&omega;}}_{\mathrm{ij}}=\frac{1}{2N}\sqrt{{(i/{\mathrm{\&theta;}}_x)}^2+{(j/{\mathrm{\&theta;}}_y)}^2},$

(2) intensity modifier value F _lumfor:

$F_{\mathrm{Lum}}=\left\{\begin{array}{cc}(60-\stackrel{\&OverBar;}{I})/150+1& \stackrel{\&OverBar;}{I}\&le;60\\ 1& 60<\stackrel{\&OverBar;}{I}<170\\ (\stackrel{\&OverBar;}{I}-170)/425+1& I\&GreaterEqual;170\end{array}\right.$

In formula, represent the mean flow rate of this macro block.

(3) contrast correction value F _contrastfor:

$F_{\mathrm{Contrast}}=\left\{\begin{array}{c}\mathrm{\&psi;},\mathrm{for}(i^2+j^2)\&le;16\mathrm{in\; Plane\; and\; Edgeblock}\\ \mathrm{\&psi;}\&CenterDot;\min (4,\max (1,{\left(\frac{C(n,i,j)}{T_{\mathrm{Basic}}(n,i,j)\&CenterDot;F_{\mathrm{Lum}}\left(n\right)}\right)}^{0.36})),\mathrm{others}\end{array}\right.$

For the macro block F that plane, edge also have texture tri-kinds dissimilar _contrastaccording to above-mentioned formulae discovery.First, candy operator is used to find out marginal points all in image; Then, edge calculation dot density function ρ _edge=Σ edge/N ²; Then according to the value of two formulae discovery ψ below:

$\mathrm{Block}\_\mathrm{type}=\left\{\begin{array}{cc}\mathrm{Plane},& {\mathrm{\&rho;}}_{\mathrm{edge}}\&le;0.1\\ \mathrm{Edge},& 0.1<{\mathrm{\&rho;}}_{\mathrm{edge}}\&le;0.2\\ \mathrm{Texture},& {\mathrm{\&rho;}}_{\mathrm{edge}}>0.2\end{array}\right.,\mathrm{\&psi;}=\left\{\begin{array}{ccc}1,& \mathrm{for}& \mathrm{Plane\; and\; Edge\; block}\\ 2.25,& \mathrm{for}& (i^2+j^2)\&le;16\mathrm{in\; Texture\; block}\\ 1.25,& \mathrm{for}& (i^2+j^2)>16\mathrm{in\; Texture\; block}\end{array}\right.$

So just can obtain the contrast correction factor.

(4) time-domain correction value F _temporalfor:

F in above formula _ttemporal frequency, f _sit is spatial frequency.

(5) marking area correction value F _foveafor:

$F_{\mathrm{Fovea}}(k,n,i,j,v,e)=W_f^{\mathrm{\&kappa;}\left(\mathrm{bg}(k,i,j)\right)}(v,e)$

In above formula, κ (bg (k, i, j)) is background luminance function, and this background luminance function is as follows:

$\mathrm{\&kappa;}\left(\mathrm{bg}(k,i,j)\right)=0.5+\frac{1}{\sqrt{2\mathrm{\&pi;}}\mathrm{\&sigma;}}\exp (-\frac{{({\log }_2(\mathrm{bg}(k,i,j)+1)-\mathrm{\&mu;})}^2}{{2\mathrm{\&sigma;}}^2})\mathrm{\&mu;}=7,\mathrm{\&sigma;}=0.8$

$W_f(v,e)=1+{(1-\frac{f_m(v,e)}{f_m(v,0)})}^{\mathrm{\&epsiv;}},\mathrm{\&epsiv;}=1.0$

f _m(v,e)＝min(f _c(e),f _d(v))

F _cdetermined by contrast sensitivity function, be set to maximum 1.0 herein; f _dbe display cut-off frequency, be set to the half of monitor resolution.Parameter e represents the distance of the saliency regional center that this position obtains to Part I.

Step 3, according to view-based access control model salient region just can examine distortion threshold calculate based on the perceived quality just can examining distortion model.

The present invention uses the calculating replacing distortion in original coding framework based on the perceived quality evaluation BMMF that just can examine distortion model, and this perceived quality evaluation BMMF is calculated as follows and obtains:

${\mathrm{BMMF}}_x\left(n\right)={\mathrm{\&Sigma;}}_{n=1}^3{w}_{\mathrm{xy}}\&CenterDot;{\hat{Q}}_{\mathrm{xy}}(B_y^o,B_y^d)$

Wherein x is the index of judgement matrix, namely belongs to which macro block; Y represents the type belonging to this macro block, and Class1 represents smooth, and type 2 represents edge, and type 3 represents texture, with represent reference macroblock respectively and rebuild macro block; Q _xyrepresent the value of x macro block under y type evaluation criterion;

${\mathrm{BMMF}}_x\left(n\right)={\mathrm{\&Sigma;}}_{n=1}^3{w}_{\mathrm{xy}}\&CenterDot;{\hat{Q}}_{\mathrm{xy}}(B_y^o,B_y^d)$

Q in above-mentioned formula is based on the perceived quality just can examining distortion model (perceived quality decision rule), and this perceived quality is obtained by PSNR-HA and PSNR-HMA two kinds of distortion criterion weightings; then represent that distortion evaluation of estimate obtains weighted average, w _xyit is weight.For PSNR-HA and PSNR-HMA, this method has done following correction on the existing basis just can examining distortion model, and concrete modification method comprises following process:

(1) for given reference block A and distortion block B, the difference of both calculating with the mean value of reference block A and distortion block B coefficient respectively;

(2) correction matrix C=B+Delt is obtained;

(3) correction factor is calculated $\mathrm{\&rho;}=\frac{\mathrm{\&Sigma;}(A-\stackrel{\&OverBar;}{A})(C-\stackrel{\&OverBar;}{C})}{\mathrm{\&Sigma;}{(C-\stackrel{\&OverBar;}{C})}^2};$

(4) revised macro block D=C × ρ is calculated;

(5) calculating just can examine the revised distortion value MSE of distortion model _hVS, as follows:

Wherein coeff _o(i, j) and coeff _d(i, j) represents the pixel value of reference block and reconstructed block correspondence position respectively.What jnd (i, j) then represented second correspondence position calculated just can examine distortion threshold, and the distortion being greater than this thresholding is that human eye can be examined, and the distortion being less than this thresholding to be human eye imperceptible, at utmost can excavate perception redundancy like this;

(6) if M ₁> M ₂then

$M_1=M_2+\left\{\begin{array}{c}(M_1-M_2)\mathrm{coef}1,\mathrm{\&rho;}<1\\ (M_1-M_2)\mathrm{coef}2,\mathrm{\&rho;}\&GreaterEqual;1\end{array}\right.;$ Finally obtain

MSE _HVS＝M ₁+Delt ²×coef3

In above-mentioned two formula, coef1, coef2, coef3 represent the modifying factor to perceptual error, get 0.002,0.25 and 0.04 respectively according to experiment experience;

(7) PSNR-HA can be obtained according to the above-mentioned perceived quality distortion calculated, as follows:

$\mathrm{PSNR}-\mathrm{HA}=10{\log }_{10}\left(\frac{{255}^2}{M}\right)$

For coloured image, M is Y, Cr, Cb luminance component, the average distortion that the perceptual distortion weighting of two color components obtains, as shown in the formula calculating:

M＝(M _Y+M _Cb×coef4+M _Cr×coef4)/(1+2×coef4)

M _y, M _cb, M _crbe a luminance component Y respectively, the perceptual distortion of two color components Cb, Cr, weight coefficient coef4 is 0.5.

Correction and the PSNR-HA of PSNR-HMA are similar, are all at calculating MSE _hVStime do corresponding correction with distortion model just can be examined, do not repeating at this.

Step 4, basis carry out rate-distortion optimization based on the perceived quality just can examining distortion model, find best coding mode.

Traditional rate-distortion optimization model is based on R-D model, and in HEVC, have employed R-λ model, and it is more accurate than R-D model.Just can examine perceived quality based on salient region of image incorporate wherein by what obtain in step 3 on the basis of R-λ model below, more be met the model of subjective quality, as follows:

$\frac{\mathrm{dJ}}{\mathrm{dR}}=\frac{\mathrm{dQ}}{\mathrm{dR}}-\mathrm{\&lambda;}=0$

In above formula, J is cost function, then λ=dQ/dR.For each macro block, calculation cost function:

j ₁＝q ₁(QP ₁)-λ·r ₁(QP ₁)for?coding?block#1

j ₂＝q ₂(QP ₂)-λ·r ₂(QP ₂)for?coding?block#2

……

j _N＝q _N(QP _N)-λ·r _N(QP _N)for?coding?block#N

$\max \left\{J\right\}=\max ({\mathrm{\&Sigma;}}_{n=1}^N{q}_n\left({\mathrm{QP}}_n\right)-\mathrm{\&lambda;}\&CenterDot;{\mathrm{\&Sigma;}}_{n=1}^N{r}_i\left({\mathrm{QP}}_n\right))$

In above formula, QP is quantization step, and the target of optimization compares by searching for seemingly the coding mode obtaining making max{J} minimum.

Do a test according to method of the present invention below, experiment effect of the present invention is described.

Test environment: Visual Studio2010;

Cycle tests: select the video test sequence of three kinds of sizes as follows from HEVC official cycle tests:

832x480：BQMall，Basketball-Drill

1280x720：Johnny，FourPeople

1920x1080：Basketball-Drive，BQterrace

Test result is as follows:

Table one Y-PSNR (dB)

Table two code check (kbps)

Table three subjective testing standard

Table four subjective test results

Experiment conclusion

This experimental result chooses six HEVC standard cycle testss.Can to find out compared with HEVC reference software HM10.0 under the identical quantization parameter of same sequence (QP) value that Y-PSNR (PSNR) will low 1.13dB ~ 4.12dB according to table one, this can tolerate more noise under prerequisite that this technology remains unchanged at Subjective video quality is described.Under can obtaining identical QP value according to table two, compare original HEVC and use the present invention can obtain less bit rate, improve compression ratio.

In subjective test, original HM10.0 compressed video is placed on left side, compressed video of the present invention is put in right side, asks 30 observers to test according to the standard of table three, obtains the result of table four after statistical average.Can find out that compressed video of the present invention is generally better than HEVC primitive technology, especially under the condition of larger QP value qualitatively supervisor; And when little QP value because the distortion after primitive technology compression is very little, so the two subjective difference can be ignored substantially.

Through objective and subjective experiment, test result all shows compared with original HEVC coding techniques, the present invention can overcome mean square error MSE as the deficiency weighing video distortion evaluation criterion, more noise is tolerated under the prerequisite that subjective quality does not reduce, remove unnecessary perception redundancy, thus raising compression efficiency, reduce the code check of the rear file of coding.

It is emphasized that; embodiment of the present invention is illustrative; instead of it is determinate; therefore the present invention includes the embodiment be not limited to described in embodiment; every other execution modes drawn by those skilled in the art's technical scheme according to the present invention, belong to the scope of protection of the invention equally.

Claims (6)

1., based on the HEVC rate-distortion optimization algorithm significantly scheming and just can examine perceived quality decision rule, it is characterized in that comprising the following steps:

Step 1, before the coding side of efficient video codec carries out mode adjudging, analyze the motor pattern of each macro block in each frame and static textural characteristics, obtain the perceived quality type of current macro, and obtain salient region of image according to different motion states;

Step 2, to calculate view-based access control model salient region according to specific image region just can examine distortion threshold;

Step 3, according to view-based access control model salient region just can examine distortion threshold calculate based on the perceived quality just can examining distortion model;

Step 4, basis carry out rate-distortion optimization based on the perceived quality just can examining distortion model.

2. the HEVC rate-distortion optimization algorithm based on just examining perceived quality decision rule according to claim 1, is characterized in that: described step 1 perceived quality type adopts following Mathematical Modeling to obtain:

${\mathrm{mp}}_k^t=\left\{\begin{array}{ccc}\mathrm{Normal\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{0,k}^t,s_k^t\&NotEqual;\mathrm{SS}\\ \mathrm{Aliased\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{1,k}^t,s_k^t\&NotEqual;\mathrm{SS}\\ \mathrm{Hysteresis\; Pattern}& \mathrm{if}& \max \left\{P_k^t\right\}\&NotEqual;p_{2,k}^t,s_k^t=\mathrm{SS}\\ \mathrm{Background}& \mathrm{if}& \max \left\{P_k^t\right\}=p_{2,k}^t,s_k^t\&NotEqual;\mathrm{SS}\end{array}\right.$

In formula, for the perceived quality type of current macro, Normal Pattern is normal perceived quality type, and Aliased Pattern is distortion-aware quality type, and Hysteresis Pattern is delayed perceived quality type, and Background is static perceived quality type, with be respectively the probability of normal condition, distortion status, inactive state appearance, for with vector form, for status attribute descriptor, SS represents inactive state.

3. the HEVC rate-distortion optimization algorithm based on just examining perceived quality decision rule according to claim 2, is characterized in that: described salient region of image is the region that the macro block of Normal Pattern type and the macro block of Hysteresis Pattern type combine formation.

4. the HEVC rate-distortion optimization algorithm based on just examining perceived quality decision rule according to claim 1, is characterized in that: the distortion threshold of just can examining of described step 2 view-based access control model salient region calculates according to the following equation:

In above-mentioned formula, FJND be view-based access control model salient region just can examine distortion threshold, T _basic(k, n, i, j), F _lum, F _contrast, F _temporaland F _foveabasic threshold value, intensity modifier value, contrast correction value, time-domain correction value and marking area correction value respectively.

5. the HEVC rate-distortion optimization algorithm based on perceived quality decision rule just can be examined according to claim 1, it is characterized in that: described step 3 is obtained by PSNR-HA and PSNR-HMA two kinds of distortion criterion weightings based on the perceived quality just can examining distortion model, and its computational methods are as follows:

(1) for given reference block A and distortion block B, the difference of both calculating with the mean value of reference block A and distortion block B coefficient respectively;

(2) correction matrix C=B+Delt is obtained;

(3) correction factor is calculated $\mathrm{\&rho;}=\frac{\mathrm{\&Sigma;}(A-\stackrel{\&OverBar;}{A})(C-\stackrel{\&OverBar;}{C})}{\mathrm{\&Sigma;}{(C-\stackrel{\&OverBar;}{C})}^2};$

(4) revised macro block D=C × ρ is calculated;

(5) calculating just can examine the revised distortion value MSE of distortion model _hVS, computational methods are as follows:

Wherein coeff _o(i, j) and coeff _d(i, j) represents the pixel value of reference block and reconstructed block correspondence position respectively, and what jnd (i, j) then represented second correspondence position calculated just can examine distortion threshold;

(6) if M ₁> M ₂then

$M_1=M_2+\left\{\begin{array}{c}(M_1-M_2)\mathrm{coef}1,\mathrm{\&rho;}<1\\ (M_1-M_2)\mathrm{coef}2,\mathrm{\&rho;}\&GreaterEqual;1\end{array}\right.;$ Finally obtain

MSE _HVS＝M ₁+Delt ²×coef3

In above formula, coef1, coef2, coef3 represent the modifying factor to perceptual error, get 0.002,0.25 and 0.04 respectively according to experiment experience;

(7) PSNR-HA is obtained according to the above-mentioned perceived quality distortion calculated:

$\mathrm{PSNR}-\mathrm{HA}=10{\log }_{10}\left(\frac{{255}^2}{M}\right)$

For coloured image, M is Y, Cr, Cb luminance component, the average distortion that the perceptual distortion weighting of two color components obtains, as shown in the formula calculating:

M＝(M _Y+M _Cb×coef4+M _Cr×coef4)/(1+2×coef4)

M _y, M _cb, M _cra luminance component Y respectively, the perceptual distortion of two color components Cb, Cr, weight coefficient coef4 is 0.5;

The correction of PSNR-HMA is at calculating MSE _hVStime do corresponding correction with distortion model just can be examined.

6. the HEVC rate-distortion optimization algorithm based on perceived quality decision rule just can be examined according to claim 1, it is characterized in that: the method that described step 4 carries out rate-distortion optimization is: just will can examine perceived quality based on salient region of image and melt and be combined on R-λ model, and obtain as drag:

$\frac{\mathrm{dJ}}{\mathrm{dR}}=\frac{\mathrm{dQ}}{\mathrm{dR}}-\mathrm{\&lambda;}=0$

In above formula, J is cost function, then λ=dQ/dR, for each macro block, and calculation cost function:

j ₁＝q ₁(QP ₁)-λ·r ₁(QP ₁)for?coding?block#1

j ₂＝q ₂(QP ₂)-λ·r ₂(QP ₂)for?coding?block#2

……

j _N＝q _N(QP _N)-λ·r _N(QP _N)for?coding?block#N

$\max \left\{J\right\}=\max ({\mathrm{\&Sigma;}}_{n=1}^N{q}_n\left({\mathrm{QP}}_n\right)-\mathrm{\&lambda;}\&CenterDot;{\mathrm{\&Sigma;}}_{n=1}^N{r}_i\left({\mathrm{QP}}_n\right))$

In above formula, QP is quantization step, compares by search the coding mode obtaining making max{J} minimum.