CN102708568B - Stereoscopic image objective quality evaluation method on basis of structural distortion - Google Patents

Abstract

The invention discloses a method for evaluating the objective quality of stereoscopic images based on structural distortion. Firstly, the undistorted stereoscopic image and the left and right viewpoint images of the distorted stereoscopic image are divided into regions to obtain human eye sensitive regions and corresponding non-sensitive regions. , and then obtain the evaluation index of sensitive area and non-sensitive area from the two aspects of structural amplitude distortion and structural direction distortion respectively; secondly, obtain the image quality evaluation value of the left and right viewpoints; re-sample the singular value difference and the mean value of the residual image after depriving the singular value The deviation rate is used to measure the distortion of the depth perception of the stereoscopic image, thereby obtaining the evaluation value of the stereoscopic perception quality; finally combining the left and right viewpoint image quality and the stereoscopic perception quality to obtain the final quality evaluation result of the stereoscopic image, because the method of the present invention avoids simulating human eyes The various components of the visual system make full use of the structural information of the stereo image, thus effectively improving the consistency between the objective evaluation results and the subjective perception.

Description

A Stereoscopic Image Objective Quality Evaluation Method Based on Structural Distortion

technical field

The invention relates to an image quality evaluation technology, in particular to an objective quality evaluation method of a stereoscopic image based on structural distortion.

Background technique

Stereoscopic image quality evaluation occupies a very important position in stereoscopic image/video systems. It can not only judge the pros and cons of processing algorithms in stereoscopic image/video systems, but also optimize and design the algorithms to improve stereoscopic image/video processing systems. s efficiency. Stereoscopic image quality assessment methods are mainly divided into two categories: subjective quality assessment and objective quality assessment. The subjective quality evaluation method is to carry out a weighted average comprehensive evaluation of the quality of the three-dimensional image to be evaluated by multiple observers. The system is difficult, so it cannot be widely promoted in practical applications. The objective quality evaluation method has the characteristics of simple operation, low cost, easy implementation and real-time optimization algorithm, etc., and has become the focus of research on stereoscopic image quality evaluation.

At present, the mainstream stereoscopic image objective quality evaluation model includes two parts: left and right viewpoint image quality evaluation and depth perception quality evaluation. However, due to the limited understanding of the human visual system, it is difficult to accurately simulate the various components of the human eye, so the agreement between these models and subjective perception is not very good.

Contents of the invention

The technical problem to be solved by the present invention is to provide an objective quality evaluation method for stereoscopic images based on structural distortion, which can effectively improve the consistency between the objective quality evaluation results and subjective perception of stereoscopic images.

The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a method for evaluating the objective quality of stereoscopic images based on structural distortion, which is characterized in that it includes the following steps:

①Let S _org be the original undistorted stereo image, let S _dis be the distorted stereo image to be evaluated, denote the left-view grayscale image of the original undistorted stereo image S _org as L _org , and denote the original undistorted stereo image The right viewpoint grayscale image of the distorted stereo image S _org is denoted as R _org , the left viewpoint grayscale image of the distorted stereo image S _dis to be evaluated is denoted as L _dis , and the right viewpoint grayscale image of the distorted stereo image S _dis to be evaluated is The viewpoint grayscale image is denoted as R _dis ;

②The four images of L _org and L _dis , R _org and R _dis are divided into regions respectively, and the corresponding sensitive area matrix maps of the four images of L _org and L _dis , R _org and R _dis are respectively obtained, and L _org and The coefficient matrices of the corresponding sensitive area matrix maps obtained after L _dis respectively implement area division are all recorded as _AL , and the coefficient value at the coordinate position (i, j) in _AL is recorded as _AL ( i, j), The coefficient matrices of the corresponding sensitive area matrix maps obtained after R _org and R _dis are divided into regions are recorded as A _R , and the coefficient value at the coordinate position (i, j) in _AR is recorded as is A _R (i,j), Among them, here 0≤i≤(W-8), 0≤j≤(H-8), W represents the width of L _org , L _dis , R _org and R _dis , H represents L _org , L _dis , R _org and the height of R _dis ;

③ Divide the two images of L _org and L _dis into (W-7)×(H-7) overlapping blocks with a size of 8×8, and then calculate that all coordinate positions in the two images of L _org and L _dis are the same The structural amplitude distortion map of two overlapping blocks of , the coefficient matrix of the structural amplitude distortion map is recorded as _BL , and the coefficient value at the coordinate position (i, j) in _BL is recorded as B _L (i, j), $B_L(i,j)=\frac{2\×{\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},L}(i,j)+C_1}{{\left({\mathrm{\σ}}_{\mathrm{org},L}(i,j)\right)}^2+{\left({\mathrm{\σ}}_{\mathrm{dis},L}(i,j)\right)}^2+C_1},$ Among them, B _L (i, j) also means that the overlapping block with the coordinate position of the upper left corner (i, j) in L _org and the size of the overlapping block of size 8×8 and the coordinate position of the upper left corner in L _dis are (i, j) Structural magnitude distortion values for overlapping blocks of size 8×8, ${\mathrm{\σ}}_{\mathrm{org},L}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(L_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},L}(i,j))}^2},$ $u_{\mathrm{org},L}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{L}_{\mathrm{org}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{dis},L}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(L_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},L}(i,j))}^2},$ $u_{\mathrm{dis},L}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{L}_{\mathrm{dis}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},L}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}((L_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},L}(i,j))\×(L_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},L}(i,j))),$ L _org (i+x, j+y) represents the pixel value of the pixel whose coordinate position is (i+x, j+y) in L _org , and L _dis (i+x, j+y) represents the coordinates in L _dis The pixel value of the pixel at the position (i+x, j+y), C ₁ represents a constant, where 0≤i≤(W-8), 0≤j≤(H-8);

Divide the two images of R _org and R _dis into (W-7)×(H-7) overlapping blocks with a size of 8×8, and then calculate the coordinates of all the same coordinates in the two images of R _org and R _dis The structural amplitude distortion map of two overlapping blocks, the coefficient matrix of the structural amplitude distortion map is recorded as _BR , and the coefficient value at the coordinate position (i,j) in _BR is recorded as B _R (i,j), $B_R(i,j)=\frac{2\×{\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},R}(i,j)+C_1}{{\left({\mathrm{\σ}}_{\mathrm{org},R}(i,j)\right)}^2+{\left({\mathrm{\σ}}_{\mathrm{dis},R}(i,j)\right)}^2+C_1},$ Among them, B _R (i, j) also means that the overlapping block with a size of 8×8 at the coordinate position of the upper left corner in R _org is (i, j) and the size at the coordinate position of the upper left corner in R _dis is (i, j) Structural magnitude distortion values for overlapping blocks of size 8×8, ${\mathrm{\σ}}_{\mathrm{org},R}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(R_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},R}(i,j))}^2},$ $u_{\mathrm{org},R}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{R}_{\mathrm{org}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{dis},R}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(R_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},R}(i,j))}^2},$ $u_{\mathrm{dis},R}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{R}_{\mathrm{dis}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},R}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}((R_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},R}(i,j))\×(R_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},R}(i,j))),$ R _org (i+x,j+y) indicates the pixel value of the pixel whose coordinate position is (i+x,j+y) in R _org , and R _dis (i+x,j+y) indicates the coordinates in R _dis The pixel value of the pixel at the position (i+x, j+y), C ₁ represents a constant, where 0≤i≤(W-8), 0≤j≤(H-8);

④ The two images of L _org and L _dis are processed by horizontal and vertical Sobel operators respectively, and the corresponding horizontal gradient matrix maps and vertical gradient matrix maps of the two images of L _org and L _dis are respectively obtained. The coefficient matrix of the corresponding horizontal gradient matrix map obtained after _org implements the Sobel operator in the horizontal direction is denoted as I _h,org,L , for the coefficient at the coordinate position (i,j) in I _h,org,L value, denoted as I _h,org,L (i,j), $\begin{array}{c}{I}_{h,\mathrm{org},L}(i,j)=L_{\mathrm{org}}(i+2,j)+2L_{\mathrm{org}}(i+2,j+1)+L_{\mathrm{org}}(i+2,j+2)\\ -L_{\mathrm{org}}(i,j)-2L_{\mathrm{org}}(i,j+1)-L_{\mathrm{org}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding gradient matrix map in the vertical direction obtained after L _org implements the Sobel operator in the vertical direction is recorded as _Iv,org,L , and the coordinate position in Iv _,org,L is (i,j) The coefficient value of , denoted as I _v,org,L (i,j), $\begin{array}{c}{I}_{v,\mathrm{org},L}(i,j)=L_{\mathrm{org}}(i,j+2)+2L_{\mathrm{org}}(i+1,j+2)+L_{\mathrm{org}}(i+2,j+2)\\ -L_{\mathrm{org}}(i,j)-2L_{\mathrm{org}}(i+1,j)-L_{\mathrm{org}}(i+2,j)\end{array},$ The coefficient matrix of the corresponding horizontal gradient matrix map obtained after L _dis is processed by the horizontal direction Sobel operator is recorded as I _{h, dis, L} , and the coordinate position in I _{h, dis, L} is (i, j) The coefficient value of , denoted as I _h,dis,L (i,j), $\begin{array}{c}{I}_{h,\mathrm{dis},L}(i,j)=L_{\mathrm{dis}}(i+2,j)+2L_{\mathrm{dis}}(i+2,j+1)+L_{\mathrm{dis}}(i+2,j+2)\\ -L_{\mathrm{dis}}(i,j)-2L_{\mathrm{dis}}(i,j+1)-L_{\mathrm{dis}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding vertical gradient matrix map obtained after L _dis is processed by the vertical direction Sobel operator is recorded as _Iv,dis,L , and the coordinate position in Iv _,dis,L is (i,j) The coefficient value of , denoted as I _v,dis,L (i,j), $\begin{array}{c}{I}_{v,\mathrm{dis},L}(i,j)=L_{\mathrm{dis}}(i,j+2)+2L_{\mathrm{dis}}(i+1,j+2)+L_{\mathrm{dis}}(i+2,j+2)\\ -L_{\mathrm{dis}}(i,j)-2L_{\mathrm{dis}}(i+1,j)-L_{\mathrm{dis}}(i+2,j)\end{array},$ Among them, L _org (i+2,j), L _org (i+2,j+1), L _org (i+2,j+2), L _org (i,j), L _org (i,j +1), L _org (i, j+2), L _org (i+1, j+2), and L _org (i+1, j) respectively indicate that the coordinate position in L _org is (i+2, j ), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1), (i,j+2), (i+1,j +2), (i+1,j) pixel value of the pixel, L _dis (i+2,j), L _dis (i+2,j+1), L _dis (i+2,j+2 ), L _dis (i,j), L _dis (i,j+1), L _dis (i,j+2), L _dis (i+1,j+2), L _dis (i+1,j ) respectively correspond to the coordinate positions in L _dis as (i+2,j), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1 ), (i, j+2), (i+1, j+2), (i+1, j) pixel values of the pixels;

The two images of R _org and R _dis are processed by horizontal and vertical Sobel operators respectively, and the corresponding horizontal gradient matrix maps and vertical gradient matrix maps of the two images of R _org and R _dis are respectively obtained, and the R _org The coefficient matrix of the corresponding gradient matrix map in the horizontal direction obtained after implementing the Sobel operator in the horizontal direction is denoted as I _h,org,R , for the coefficient value at the coordinate position (i,j) in I _h,org,R , denoted as I _h,org,R (i,j), $\begin{array}{c}{I}_{h,\mathrm{org},R}(i,j)=R_{\mathrm{org}}(i+2,j)+2R_{\mathrm{org}}(i+2,j+1)+R_{\mathrm{org}}(i+2,j+2)\\ -R_{\mathrm{org}}(i,j)-2R_{\mathrm{org}}(i,j+1)-R_{\mathrm{org}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding gradient matrix map in the vertical direction obtained after R _org implements the Sobel operator in the vertical direction is recorded as _Iv,org,R , and the coordinate position in Iv _,org,R is (i,j) The coefficient value of , denoted as I _v,org,R (i,j), $\begin{array}{c}{I}_{v,\mathrm{org},R}(i,j)=R_{\mathrm{org}}(i,j+2)+2R_{\mathrm{org}}(i+1,j+2)+R_{\mathrm{org}}(i+2,j+2)\\ -R_{\mathrm{org}}(i,j)-2R_{\mathrm{org}}(i+1,j)-R_{\mathrm{org}}(i+2,j)\end{array},$ The coefficient matrix of the corresponding horizontal gradient matrix map obtained after R _dis is processed by the horizontal direction Sobel operator is recorded as I _{h, dis, R} , and the coordinate position in I _{h, dis, R} is (i, j) The coefficient value of , denoted as I _h,dis,R (i,j), $\begin{array}{c}{I}_{h,\mathrm{dis},R}(i,j)=R_{\mathrm{dis}}(i+2,j)+2R_{\mathrm{dis}}(i+2,j+1)+R_{\mathrm{dis}}(i+2,j+2)\\ -R_{\mathrm{dis}}(i,j)-2R_{\mathrm{dis}}(i,j+1)-R_{\mathrm{dis}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding gradient matrix map in the vertical direction obtained after R _dis is processed by the vertical direction Sobel operator is recorded as _Iv,dis,R , and the coordinate position in Iv _,dis,R is (i,j) The coefficient value of , denoted as I _v,dis,R (i,j), $\begin{array}{c}{I}_{v,\mathrm{dis},R}(i,j)=R_{\mathrm{dis}}(i,j+2)+2R_{\mathrm{dis}}(i+1,j+2)+R_{\mathrm{dis}}(i+2,j+2)\\ -R_{\mathrm{dis}}(i,j)-2R_{\mathrm{dis}}(i+1,j)-R_{\mathrm{dis}}(i+2,j)\end{array},$ Among them, R _org (i+2,j), R _org (i+2,j+1), R _org (i+2,j+2), R _org (i,j), R _org (i,j +1), R _org (i, j+2), R _org (i+1, j+2), and R _org (i+1, j) respectively indicate that the coordinate position in R _org is (i+2, j ), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1), (i,j+2), (i+1,j +2), (i+1,j) pixel value of the pixel, R _dis (i+2,j), R _dis (i+2,j+1), R _dis (i+2,j+2 ), R _dis (i,j), R _dis (i,j+1), R _dis (i,j+2), R _dis (i+1,j+2), R _dis (i+1,j ) respectively correspond to the coordinate positions in R _dis as (i+2,j), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1 ), (i, j+2), (i+1, j+2), (i+1, j) the pixel value of the pixel point;

⑤ Calculate the structural direction distortion map of two overlapping blocks with the same coordinate position in the two images of L _org and L _dis , record the coefficient matrix of the structural direction distortion map as E _L , and the coordinate position in E _L is ( i,j), denote it as E _L (i,j), ${\mathrm{E.}}_L(i,j)=\frac{I_{h,\mathrm{org},L}(i,j)\×I_{h,\mathrm{dis},L}(i,j)+I_{v,\mathrm{org},L}(i,j)\×I_{v,\mathrm{dis},L}(i,j)+C_2}{\sqrt{{\left(I_{h,\mathrm{org},L}(i,j)\right)}^2+{\left(I_{v,\mathrm{org},L}(i,j)\right)}^2}\×\sqrt{{\left(I_{h,\mathrm{dis},L}(i,j)\right)}^2+{\left(I_{v,\mathrm{dis},L}(i,j)\right)}^2}+C_2},$ Wherein, C represents a _constant ;

Calculate the structural direction distortion map of two overlapping blocks with the same coordinate position in the two images of R _org and R _dis , record the coefficient matrix of the structural direction distortion map as E _R , and the coordinate position in E _R is (i ,j), which is recorded as E _R (i,j), ${\mathrm{E.}}_R(i,j)=\frac{I_{h,\mathrm{org},R}(i,j)\×I_{h,\mathrm{dis},R}(i,j)+I_{v,\mathrm{org},R}(i,j)\×I_{v,\mathrm{dis},R}(i,j)+C_2}{\sqrt{{\left(I_{h,\mathrm{org},R}(i,j)\right)}^2+{\left(I_{v,\mathrm{org},R}(i,j)\right)}^2}\×\sqrt{{\left(I_{h,\mathrm{dis},R}(i,j)\right)}^2+{\left(I_{v,\mathrm{dis},R}(i,j)\right)}^2}+C_2};$

⑥ Calculate the structural distortion evaluation value of L _org and L _dis , denoted as Q ^L , Q ^L ＝ω ₁ ×Q _m,L +ω ₂ ×Q _nm,L , $Q_{m,L}=\frac{1}{N_{L,m}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_L(i,j)+{\mathrm{E.}}_L(i,j))\×A_L(i,j)),$ $N_{L,m}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}{A}_L(i,j),$ $Q_{\mathrm{nm},L}=\frac{1}{N_{L,\mathrm{nm}}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_L(i,j)+{\mathrm{E.}}_L(i,j))\×(1-A_L(i,j))),$ $N_{L,\mathrm{nm}}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(1-A_L(i,j)),$ Among them, ω ₁ represents the weight value of the sensitive area in L _org and L _dis , ω ₂ represents the weight value of the non-sensitive area in L _org and L _dis ;

Calculate the structural distortion evaluation value of R _org and R _dis , denoted as Q ^R , Q ^R ＝ω' ₁ ×Q _m,R +ω' ₂ ×Q _nm,R , $Q_{m,R}=\frac{1}{N_{R,m}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_R(i,j)+{\mathrm{E.}}_R(i,j))\×A_R(i,j)),$ $N_{R,m}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}{A}_R(i,j),$ $Q_{\mathrm{nm},R}=\frac{1}{N_{R,\mathrm{nm}}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_R(i,j)+{\mathrm{E.}}_R(i,j))\×(1-A_R(i,j))),$ $N_{R,\mathrm{nm}}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(1-A_R(i,j)),$ Among them, _ω'1 represents the weight value of the sensitive area in R _org and R _dis , and _ω'2 represents the weight value of the non-sensitive area in R _org and R _dis ;

⑦ Calculate the spatial frequency similarity measure of the distorted stereo image S _dis to be evaluated relative to the original undistorted stereo image S _org according to Q ^L and Q ^R , denoted as Q _F , Q _F =β ₁ ×Q ^L +( 1-β ₁ )×Q ^R , where β ₁ represents the weight of Q ^L ;

⑧ Calculate the absolute difference image of L _org and R _org , expressed in matrix form as Calculate the absolute difference image of L _dis and R _dis , expressed in matrix form as Among them, "||" is the absolute value symbol;

⑨ will and The image is divided into non-overlapping blocks of size 8×8, and then and Singular value decomposition is performed on all the blocks in the image respectively to get The corresponding singular value map composed of the singular value matrix of each block and The corresponding singular value map composed of the singular value matrix of each block will be The coefficient matrix of the singular value map obtained after implementing singular value decomposition is denoted as G _org , and for the singular value at the coordinate position (p, q) in the singular value matrix of the nth block in G _org , it is denoted as Will The coefficient matrix of the singular value map obtained after the implementation of singular value decomposition is denoted as G _dis , and for the singular value at the coordinate position (p, q) in the singular value matrix of the nth block in G _dis , it is denoted as Among them, W _LR means and The width, H _LR means and height of, 0≤p≤7,0≤q≤7;

⑩ Calculation The corresponding singular value maps and The singular value deviation evaluation value of the corresponding singular value map, denoted as K, $K=\frac{64}{W_{\mathrm{LR}}\×h_{\mathrm{LR}}}\×\underset{\mathrm{no}=0}{\overset{W_{\mathrm{LR}}\×h_{\mathrm{LR}}/64-1}{\mathrm{\Σ}}}\frac{\underset{p=0}{\overset{7}{\mathrm{\Σ}}}(G_{\mathrm{org}}^{\mathrm{no}}(p,p)\×|G_{\mathrm{org}}^{\mathrm{no}}(p,p)-G_{\mathrm{dis}}^{\mathrm{no}}(p,p)\left|\right)}{\underset{p=0}{\overset{7}{\mathrm{\Σ}}}{G}_{\mathrm{org}}^{\mathrm{no}}(p,p)},$ in, Indicates the singular value at the coordinate position (p,p) in the singular value matrix of the nth block in G _org , Represents the singular value at the coordinate position (p, p) in the singular value matrix of the nth block in G _dis ;

right and Singular value decomposition is carried out respectively, and we get and Corresponding to two orthogonal matrices and one singular value matrix, the The two orthogonal matrices obtained after implementing singular value decomposition are denoted as χ _org and V _org respectively, and the The singular value matrix obtained after implementing singular value decomposition is denoted as O _org , Will The two orthogonal matrices obtained after implementing singular value decomposition are denoted as χ _dis and V _dis respectively, and the The singular value matrix obtained after implementing singular value decomposition is denoted as O _dis ,

Calculate separately and The residual matrix image after stripping the singular values of the images, the The residual matrix after deprivation of singular values is denoted as X _org , X _org = χ _org × Λ × V _org , the The residual matrix after stripping the singular value is marked as X _dis , X _dis =χ _dis ×Λ×V _dis , where Λ represents the identity matrix, and the size of Λ is consistent with the size of O _org and O _dis ;

Calculate the mean deviation rate of X _org and X _dis , denoted as Wherein, x represents the abscissa of the pixel in X _org and X _dis , and y represents the ordinate of the pixel in X _org and X _dis ;

Calculate the stereo perception evaluation metric of the distorted stereo image S _dis to be evaluated relative to the original undistorted stereo image S _org , denoted as Q _S , where τ represents a constant used to adjust K and The importance played in _QS ;

According to Q _F and Q _S , calculate the image quality evaluation score of the distorted stereo image S _dis to be evaluated, denoted as Q, Q=Q _F ×(Q _S ) ^ρ , where ρ represents the weight coefficient value.

In the step 2., the acquisition process of the coefficient matrix _AL of the corresponding sensitive area matrix map of L _org and L _dis is:

②-a1. Perform horizontal and vertical Sobel operator processing on L _org to obtain the horizontal gradient image and vertical gradient image of L _org , which are recorded as Z _h,l1 and Z _v,l1 respectively, and then calculate the gradient of L _org Amplitude diagram, denoted as Z _l1 , Among them, Z _l1 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _l1 , Z _{h, l1} (x, y) represents the coordinate position in Z _{h, l1} is (x, y) the horizontal direction gradient value of the pixel point, Z _v,l1 (x,y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x,y) in Z _v,l1 , 1≤x≤W', 1≤y≤H', where W' represents the width of Z _l1 , and H' represents the height of Z _l1 ;

②-a2. Perform horizontal and vertical Sobel operator processing on L _dis to obtain the horizontal gradient image and vertical gradient image of L _dis , which are recorded as Z _h,l2 and Z _v,l2 respectively, and then calculate the gradient of L _dis Amplitude diagram, denoted as Z _l2 , Among them, Z _l2 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _l2 , Z _{h, l2} (x, y) represents the coordinate position in Z _{h, l2} is (x, y) the horizontal direction gradient value of the pixel point, Z _v,l2 (x,y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x,y) in Z _v,l2 , 1≤x≤W', 1≤y≤H', where W' represents the width of Z _l2 , and H' represents the height of Z _l2 ;

②-a3. Calculate the threshold T required for dividing the area, $T=\mathrm{\α}\×\frac{1}{W^{\′}\×h^{\′}}\×(\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{l1}(x,\mathrm{the\; y})+\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{l2}(x,\mathrm{the\; y})),$ Among them, α is a constant, Z _l1 (x, y) represents the gradient magnitude of the pixel point whose coordinate position in Z _l1 is (x, y), Z _l2 (x, y) represents the coordinate position in Z _l2 is (x, y) The gradient magnitude of the pixel point of y);

②-a4. Record the gradient magnitude of the pixel at the coordinate position (i, j) in Z _l1 as Z _l1 (i, j), and the gradient of the pixel at the coordinate position (i, j) in Z _l2 The amplitude is recorded as Z _l2 (i,j), and it is judged whether Z _l1 (i,j)>T or Z _l2 (i,j)>T is established. If it is established, the coordinate position in L _org and L _dis is determined as ( i, j) belongs to the sensitive area, and let A _L (i, j) = 1, otherwise, determine that the pixel with the coordinate position (i, j) in L _org and L _dis belongs to the non-sensitive area, and set A _L (i,j)=0, where, 0≤i≤(W-8), 0≤j≤(H-8);

The acquisition process of the coefficient matrix A _R of the sensitive area matrix map corresponding to R _org and R _dis respectively in the step ② is:

②-b1. Perform horizontal and vertical Sobel operator processing on R _org to obtain the horizontal gradient image and vertical gradient image of R _org , which are recorded as Z _h,r1 and Z _v,r1 respectively, and then calculate the gradient of R _org Amplitude diagram, denoted as Z _r1 , Among them, Z _r1 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _r1 , Z _{h, r1} (x, y) represents the coordinate position in Z _{h, r1} is (x, The horizontal direction gradient value of the pixel point of y), Z _{v, r1} (x, y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x, y) in Z _{v, r1} , 1≤x≤W', 1≤y≤H', where W' indicates the width of Z _r1 , and H' indicates the height of Z _r1 ;

②-b2. Perform horizontal and vertical Sobel operator processing on R _dis to obtain the horizontal gradient image and vertical gradient image of R _dis , which are recorded as Z _h,r2 and Z _v,r2 respectively, and then calculate the gradient of R _dis Magnitude diagram, denoted as Z _r2 , Among them, Z _r2 (x, y) represents the gradient magnitude of the pixel whose coordinate position in Z _r2 is (x, y), Z _{h, r2} (x, y) represents the coordinate position in Z _{h, r2} is (x, The horizontal direction gradient value of the pixel point of y), Z _{v, r2} (x, y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x, y) in Z _{v, r2} , 1≤x≤W', 1≤y≤H', where W' indicates the width of Z _r2 , and H' indicates the height of Z _r2 ;

②-b3. Calculate the threshold T' required for dividing the area, $T^{\′}=\mathrm{\α}\×\frac{1}{W^{\′}\×h^{\′}}\×(\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{r1}(x,\mathrm{the\; y})+\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{r2}(x,\mathrm{the\; y})),$ Among them, α is a constant, Z _r1 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _r1 , Z _r2 (x, y) represents the coordinate position in Z _r2 is (x, y) The gradient magnitude of the pixel point of y);

②-b4. Record the gradient magnitude of the pixel at the coordinate position (i, j) in Z _r1 as Z _r1 (i, j), and record the gradient of the pixel at the coordinate position (i, j) in Z _r2 The amplitude is recorded as Z _r2 (i,j), judge whether Z _r1 (i,j)>T or Z _r2 (i,j)>T is true, if true, determine the coordinate position in R _org and R _dis as ( i, j) belongs to the sensitive area, and set A _R (i, j) = 1, otherwise, determine that the pixel with the coordinate position (i, j) in R _org and R _dis belongs to the non-sensitive area, and set A _R (i,j)=0, wherein, 0≤i≤(W-8), 0≤j≤(H-8).

The acquisition process of _β1 in the described step ⑦ is:

⑦-1. Using n undistorted stereoscopic images to establish a distorted stereoscopic image set under different distortion types and different degrees of distortion, the distorted stereoscopic image set includes multiple distorted stereoscopic images, wherein n≥1;

⑦-2. Utilize the subjective quality evaluation method to obtain the average subjective score difference value of each distorted stereoscopic image in the distorted stereoscopic image set, which is denoted as DMOS, DMOS=100-MOS, wherein, MOS represents the mean value of the subjective score, and DMOS∈[0,100 ];

⑦-3, according to the operation process of step ① to step ⑥, calculate the evaluation value Q of the sensitive area of the left viewpoint image of each distorted stereoscopic image in the distorted stereoscopic image set relative to the left viewpoint image of the corresponding undistorted stereoscopic image _{m, L} and the evaluation value Q _{nm, L} of the non-sensitive area;

⑦-4. Using a mathematical fitting method to fit the average subjective score difference DMOS and the corresponding Q _m,L and Q _nm,L of the distorted stereo images in the distorted stereo image set, so as to obtain the value of β ₁ .

Compared with the prior art, the present invention has the advantage of firstly dividing the left viewpoint image and the right viewpoint image of the undistorted stereoscopic image and the distorted stereoscopic image respectively to obtain human eye sensitive regions and corresponding non-sensitive regions, and then The evaluation indexes of sensitive area and non-sensitive area are respectively obtained from the two aspects of structural amplitude distortion and structural direction distortion; secondly, linear weighting is used to obtain the left viewpoint image quality evaluation value and right viewpoint image quality evaluation value respectively, and then the left and right viewpoint image quality is obtained Evaluation value; again according to the singular value can better characterize the structural information characteristics of the stereoscopic image, the difference of the sampling singular value and the mean deviation rate of the residual image after depriving the singular value are used to measure the distortion of the depth perception of the stereoscopic image, so as to obtain the stereoscopic perception quality evaluation value; finally the left and right viewpoint image quality and the stereoscopic perception quality are combined in a non-linear manner to obtain the final quality evaluation result of the stereoscopic image. Since the method of the present invention avoids simulating each component part of the human visual system, fully utilizes Structural information of stereoscopic images, thus effectively improving the consistency of objective evaluation results with subjective perception.

Description of drawings

Fig. 1 is the overall realization block diagram of the inventive method;

Figure 2a is a stereoscopic image of Akko&Kayo (640×480);

Figure 2b is a stereoscopic image of Alt Moabit (1024×768);

Figure 2c is a stereoscopic image of Balloons (1024×768);

Figure 2d is a stereoscopic image of Door Flowers (1024×768);

Figure 2e is a stereoscopic image of Kendo (1024×768);

Figure 2f is a stereo image of Leaving Laptop (1024×768);

Figure 2g is a stereoscopic image of Lovebird1 (1024×768);

Figure 2h is a stereoscopic image of Newspaper (1024×768);

Fig. 2i is Xmas (640 * 480) stereoscopic image;

Fig. 2j is Puppy (720 * 480) three-dimensional image;

Fig. 2k is Soccer2 (720 * 480) three-dimensional image;

Figure 2l is a stereoscopic image of Horse (480×270);

Fig. 3 is a block diagram of left viewpoint image quality evaluation of the method of the present invention;

Figure 4a is a diagram of the CC performance change between the left view image quality and subjective perception quality under different α and ω ₁ ;

Figure 4b is a graph of the SROCC performance change between the left viewpoint image quality and subjective perception quality under different α and ω ₁ ;

Figure 4c is a graph of the RMSE performance change between the left view image quality and subjective perception quality under different α and ω ₁ ;

Fig. 5a is a diagram of CC performance change between left view image quality and subjective perception quality under different α in the case of ω ₁ =1;

Fig. 5b is a graph of SROCC performance change between left view image quality and subjective perception quality under different α in the case of ω ₁ =1;

Fig. 5c is a diagram of RMSE performance change between left view image quality and subjective perception quality under different α in the case of ω ₁ =1;

Figure 6a is a diagram of the CC performance change between left and right viewpoint image quality and subjective perception quality under different _β1 ;

Figure 6b is a graph of the SROCC performance change between left and right viewpoint image quality and subjective perception quality under different _β1 ;

Fig. 6c is the RMSE performance change diagram between left and right viewpoint image quality and subjective perception quality under different _β1 ;

Figure 7a is a graph of the CC performance change between the stereoscopic depth perception quality and the subjective perception quality under different τ;

Figure 7b is a graph of the SROCC performance change between the stereoscopic depth perception quality and the subjective perception quality under different τ;

Figure 7c is a graph of the RMSE performance change between the stereoscopic depth perception quality and the subjective perception quality under different τ;

Figure 8a is a diagram of CC performance variation between stereoscopic image quality and subjective perceptual quality under different ρ;

Figure 8b is a graph of the SROCC performance change between the stereoscopic image quality and the subjective perception quality under different ρ;

Fig. 8c is a graph of RMSE performance variation between stereoscopic image quality and subjective perception quality under different ρ.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

A method for evaluating the objective quality of a stereoscopic image based on structural distortion proposed by the present invention, which evaluates the image quality of the left and right viewpoints and the stereoscopic perception quality of the stereoscopic image from the perspective of structural distortion, and obtains the final quality of the stereoscopic image in a non-linear weighted manner Evaluation value. Fig. 1 has provided the overall realization block diagram of the inventive method, and it comprises the following steps:

① Let S _org be the original undistorted stereo image, let S _dis be the distorted stereo image to be evaluated, denote the left-viewpoint grayscale image of the original undistorted stereo image S _org as L _org , and denote the original undistorted stereo image The right viewpoint grayscale image of the distorted stereo image S _org is denoted as R _org , the left viewpoint grayscale image of the distorted stereo image S _dis to be evaluated is denoted as L _dis , and the right viewpoint grayscale image of the distorted stereo image S _dis to be evaluated is The viewpoint grayscale image is denoted as R _dis .

②The four images of L _org and L _dis , R _org and R _dis are divided into regions respectively, and the corresponding sensitive area matrix maps of the four images of L _org and L _dis , R _org and R _dis are respectively obtained, and L _org and The coefficient matrices of the corresponding sensitive area matrix maps obtained after L _dis are respectively divided into regions are all recorded as _AL , and the coefficient value at the coordinate position (i, j) in _AL is recorded as _AL ( i, j), The coefficient matrices of the corresponding sensitive area matrix maps obtained after R _org and R _dis are divided into regions are recorded as A _R , and the coefficient value at the coordinate position (i, j) in _AR is recorded as is A _R (i,j), Among them, here 0≤i≤(W-8), 0≤j≤(H-8), W represents the width of L _org , L _dis , R _org and R _dis , H represents L _org , L _dis , R _org and R _dis high.

In this specific embodiment, the acquisition process of the coefficient matrix _AL of the sensitive area matrix map corresponding to L _org and L _dis respectively in step 2. is:

②-a1. Perform horizontal and vertical Sobel operator processing on L _org to obtain the horizontal gradient image and vertical gradient image of L _org , which are recorded as Z _h,l1 and Z _v,l1 respectively, and then calculate the gradient of L _org Amplitude diagram, denoted as Z _l1 , Among them, Z _l1 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _l1 , Z _{h, l1} (x, y) represents the coordinate position in Z _{h, l1} is (x, y) the horizontal direction gradient value of the pixel point, Z _v,l1 (x,y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x,y) in Z _v,l1 , 1≤x≤W', 1≤y≤H', where W' represents the width of Z _l1 , and H' represents the height of Z _l1 .

②-a2. Perform horizontal and vertical Sobel operator processing on L _dis to obtain the horizontal gradient image and vertical gradient image of L _dis , which are recorded as Z _h,l2 and Z _v,l2 respectively, and then calculate the gradient of L _dis Amplitude diagram, denoted as Z _l2 , Among them, Z _l2 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _l2 , Z _{h, l2} (x, y) represents the coordinate position in Z _{h, l2} is (x, y) the horizontal direction gradient value of the pixel point, Z _v,l2 (x,y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x,y) in Z _v,l2 , 1≤x≤W', 1≤y≤H', where W' represents the width of Z _l2 , and H' represents the height of Z _l2 .

②-a3. Calculate the threshold T required for dividing the area, $T=\mathrm{\α}\×\frac{1}{W^{\′}\×h^{\′}}\×(\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{l1}(x,\mathrm{the\; y})+\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{l2}(x,\mathrm{the\; y})),$ Among them, W' represents the width of Z _l1 and Z _l2 , H' represents the height of Z _l1 and Z _l2 , α is a constant, Z _l1 (x, y) represents the pixel point in Z _l1 whose coordinate position is (x, y) The gradient magnitude of Z _l2 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _l2 .

②-a4. Record the gradient magnitude of the pixel point whose coordinate position is (i, j) in Z _l1 as Z _l1 (i, j), and record the gradient amplitude of the pixel point whose coordinate position is (i, j) in Z _l2 The amplitude is recorded as Z _l2 (i,j), and it is judged whether Z _l1 (i,j)>T or Z _l2 (i,j)>T is established. If it is established, the coordinate position in L _org and L _dis is determined as ( i, j) belongs to the sensitive area, and let A _L (i, j) = 1, otherwise, determine that the pixel with the coordinate position (i, j) in L _org and L _dis belongs to the non-sensitive area, and set A _L (i,j)=0, wherein, 0≤i≤(W-8), 0≤j≤(H-8).

In this specific embodiment, the acquisition process of the coefficient matrix _AR of the sensitive area matrix map corresponding to R _org and R _dis respectively in step ② is:

②-b1. Perform horizontal and vertical Sobel operator processing on R _org to obtain the horizontal gradient image and vertical gradient image of R _org , which are recorded as Z _h,r1 and Z _v,r1 respectively, and then calculate the gradient of R _org Amplitude diagram, denoted as Z _r1 , Among them, Z _r1 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _r1 , Z _{h, r1} (x, y) represents the coordinate position in Z _{h, r1} is (x, The horizontal direction gradient value of the pixel point of y), Z _{v, r1} (x, y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x, y) in Z _{v, r1} , 1≤x≤W', 1≤y≤H', where W' represents the width of Z _r1 , and H' represents the height of Z _r1 .

②-b2. Perform horizontal and vertical Sobel operator processing on R _dis to obtain the horizontal gradient image and vertical gradient image of R _dis , which are recorded as Z _h,r2 and Z _v,r2 respectively, and then calculate the gradient of R _dis Amplitude diagram, denoted as Z _r2 , Among them, Z _r2 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _r2 , Z _{h, r2} (x, y) represents the coordinate position in Z _{h, r2} is (x, The horizontal direction gradient value of the pixel point of y), Z _{v, r2} (x, y) represents the vertical direction gradient value of the pixel point whose coordinate position is (x, y) in Z _{v, r2} , 1≤x≤W', 1≤y≤H', where W' represents the width of Z _r2 , and H' represents the height of Z _r2 .

②-b3. Calculate the threshold T' required for dividing the area, $T^{\′}=\mathrm{\α}\×\frac{1}{W^{\′}\×h^{\′}}\×(\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{r1}(x,\mathrm{the\; y})+\underset{x=0}{\overset{W^{\′}}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{h^{\′}}{\mathrm{\Σ}}}{Z}_{r2}(x,\mathrm{the\; y})),$ Among them, W' represents the width of Z _r1 and Z _r2 , H' represents the height of Z _r1 and Z _r2 , α is a constant, and Z _r1 (x, y) represents the pixel in Z _r1 whose coordinate position is (x, y) The gradient magnitude of Z _r2 (x, y) represents the gradient magnitude of the pixel whose coordinate position is (x, y) in Z _r2 .

②-b4. Record the gradient magnitude of the pixel at the coordinate position (i, j) in Z _r1 as Z _r1 (i, j), and the gradient of the pixel at the coordinate position (i, j) in Z _r2 The amplitude is recorded as Z _r2 (i,j), judge whether Z _r1 (i,j)>T or Z _r2 (i,j)>T is true, if true, determine the coordinate position in R _org and R _dis as ( i, j) belongs to the sensitive area, and set A _R (i, j) = 1, otherwise, determine that the pixel with the coordinate position (i, j) in R _org and R _dis belongs to the non-sensitive area, and set A _R (i,j)=0, wherein, 0≤i≤(W-8), 0≤j≤(H-8).

In this embodiment, the 12 distortion-free Stereoscopic images of different distortion types and different degrees of distortion of the stereoscopic image set, the distortion types include JPEG compression, JP2K compression, Gaussian white noise, Gaussian blur and H264 encoding distortion, and the left viewpoint image and right viewpoint image of the stereoscopic image At the same time, the distorted stereo image set includes 312 distorted stereo images, including 60 distorted stereo images compressed by JPEG, 60 distorted stereo images compressed by JPEG2000, and 60 distorted stereo images compressed by Gaussian white noise. There are 60 images, 60 stereoscopic images with Gaussian blur distortion, and 72 stereoscopic images with H264 encoding distortion. The above-mentioned 312 stereoscopic images are divided into regions as described above.

In this embodiment, the α value determines the accuracy of the sensitive area division. If the value is too large, the sensitive area will be mistaken for a non-sensitive area; if the value is too small, the non-sensitive area will be mistaken for a sensitive area. , so the determination process of its value is determined by the contribution of left-viewpoint image quality or right-viewpoint image quality to stereoscopic image quality.

③ Divide the two images of L _org and L _dis into (W-7)×(H-7) overlapping blocks with a size of 8×8, and then calculate that all coordinate positions in the two images of L _org and L _dis are the same The structural magnitude-distortion map of two overlapping blocks of , the coefficient matrix of the structural magnitude-distortion map is recorded as _BL , and the coefficient value at the coordinate position (i, j) in _BL is recorded as B _L (i, j), $B_L(i,j)=\frac{2\×{\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},L}(i,j)+C_1}{{\left({\mathrm{\σ}}_{\mathrm{org},L}(i,j)\right)}^2+{\left({\mathrm{\σ}}_{\mathrm{dis},L}(i,j)\right)}^2+C_1},$ Among them, B _L (i, j) also means that the overlapping block with a size of 8×8 at the coordinate position of the upper left corner in L _org is (i, j) and the size at the coordinate position of the upper left corner in L _dis is (i, j) Structural magnitude distortion values for overlapping blocks of size 8×8, ${\mathrm{\σ}}_{\mathrm{org},L}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(L_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},L}(i,j))}^2},$ $u_{\mathrm{org},L}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{L}_{\mathrm{org}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{dis},L}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(L_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},L}(i,j))}^2},$ $u_{\mathrm{dis},L}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{L}_{\mathrm{dis}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},L}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}((L_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},L}(i,j))\×(L_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},L}(i,j))),$ L _org (i+x, j+y) represents the pixel value of the pixel whose coordinate position is (i+x, j+y) in L _org , and L _dis (i+x, j+y) represents the coordinates in L _dis The pixel value of the pixel at the position (i+x, j+y), C ₁ represents a constant, C ₁ is to avoid $B_L(i,j)=\frac{2\×{\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},L}(i,j)+C_1}{{\left({\mathrm{\σ}}_{\mathrm{org},L}(i,j)\right)}^2+{\left({\mathrm{\σ}}_{\mathrm{dis},L}(i,j)\right)}^2+C_1}$ In the case where the denominator of is zero, C ₁ =0.01 may be taken in practical application, where 0≤i≤(W-8), 0≤j≤(H-8).

Here, considering the correlation between the pixels of the image, an overlapping block with a size of 8×8 overlaps with its nearest left or right block by 7 columns. Similarly, the 8×8 overlapping block overlaps with its The most adjacent upper or lower blocks have 7 lines of overlap.

Divide the two images of R _org and R _dis into (W-7)×(H-7) overlapping blocks with a size of 8×8, and then calculate the coordinates of all the same coordinates in the two images of R _org and R _dis The structural amplitude distortion map of two overlapping blocks, the coefficient matrix of the structural amplitude distortion map is recorded as _BR , and the coefficient value at the coordinate position (i,j) in _BR is recorded as B _R (i,j), $B_R(i,j)=\frac{2\×{\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},R}(i,j)+C_1}{{\left({\mathrm{\σ}}_{\mathrm{org},R}(i,j)\right)}^2+{\left({\mathrm{\σ}}_{\mathrm{dis},R}(i,j)\right)}^2+C_1},$ Among them, B _R (i, j) also means that the overlapping block with a size of 8×8 at the coordinate position of the upper left corner in R _org is (i, j) and the size at the coordinate position of the upper left corner in R _dis is (i, j) Structural magnitude distortion values for overlapping blocks of size 8×8, ${\mathrm{\σ}}_{\mathrm{org},R}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(R_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},R}(i,j))}^2},$ $u_{\mathrm{org},R}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{R}_{\mathrm{org}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{dis},R}(i,j)=\sqrt{\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{(R_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},R}(i,j))}^2},$ $u_{\mathrm{dis},R}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}{R}_{\mathrm{dis}}(i+x,j+\mathrm{the\; y}),$ ${\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},R}(i,j)=\frac{1}{64}\underset{x=0}{\overset{7}{\mathrm{\Σ}}}\underset{\mathrm{the\; y}=0}{\overset{7}{\mathrm{\Σ}}}((R_{\mathrm{org}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{org},R}(i,j))\×(R_{\mathrm{dis}}(i+x,j+\mathrm{the\; y})-u_{\mathrm{dis},R}(i,j))),$ R _org (i+x,j+y) indicates the pixel value of the pixel whose coordinate position is (i+x,j+y) in R _org , and R _dis (i+x,j+y) indicates the coordinates in R _dis The pixel value of the pixel at the position (i+x, j+y), C ₁ represents a constant, C ₁ is to avoid $B_R(i,j)=\frac{2\×{\mathrm{\σ}}_{\mathrm{org},\mathrm{dis},R}(i,j)+C_1}{{\left({\mathrm{\σ}}_{\mathrm{org},R}(i,j)\right)}^2+{\left({\mathrm{\σ}}_{\mathrm{dis},R}(i,j)\right)}^2+C_1}$ In the case where the denominator of is zero, C ₁ =0.01 may be taken in practical application, where 0≤i≤(W-8), 0≤j≤(H-8).

④ The two images of L _org and L _dis are processed by horizontal and vertical Sobel operators respectively, and the corresponding horizontal gradient matrix maps and vertical gradient matrix maps of the two images of L _org and L _dis are respectively obtained. The coefficient matrix of the corresponding horizontal gradient matrix map obtained after _org implements the Sobel operator in the horizontal direction is denoted as I _h,org,L , for the coefficient at the coordinate position (i,j) in I _h,org,L value, denoted as I _h,org,L (i,j), $\begin{array}{c}{I}_{h,\mathrm{org},L}(i,j)=L_{\mathrm{org}}(i+2,j)+2L_{\mathrm{org}}(i+2,j+1)+L_{\mathrm{org}}(i+2,j+2)\\ -L_{\mathrm{org}}(i,j)-2L_{\mathrm{org}}(i,j+1)-L_{\mathrm{org}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding gradient matrix map in the vertical direction obtained after L _org implements the Sobel operator in the vertical direction is recorded as _Iv,org,L , and the coordinate position in Iv _,org,L is (i,j) The coefficient value of , denoted as I _v,org,L (i,j), $\begin{array}{c}{I}_{v,\mathrm{org},L}(i,j)=L_{\mathrm{org}}(i,j+2)+2L_{\mathrm{org}}(i+1,j+2)+L_{\mathrm{org}}(i+2,j+2)\\ -L_{\mathrm{org}}(i,j)-2L_{\mathrm{org}}(i+1,j)-L_{\mathrm{org}}(i+2,j)\end{array},$ The coefficient matrix of the corresponding horizontal gradient matrix map obtained after L _dis is processed by the horizontal direction Sobel operator is recorded as I _{h, dis, L} , and the coordinate position in I _{h, dis, L} is (i, j) The coefficient value of , denoted as I _h,dis,L (i,j), $\begin{array}{c}{I}_{h,\mathrm{dis},L}(i,j)=L_{\mathrm{dis}}(i+2,j)+2L_{\mathrm{dis}}(i+2,j+1)+L_{\mathrm{dis}}(i+2,j+2)\\ -L_{\mathrm{dis}}(i,j)-2L_{\mathrm{dis}}(i,j+1)-L_{\mathrm{dis}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding vertical gradient matrix map obtained after L _dis is processed by the vertical direction Sobel operator is recorded as _Iv,dis,L , and the coordinate position in Iv _,dis,L is (i,j) The coefficient value of , denoted as I _v,dis,L (i,j), $\begin{array}{c}{I}_{v,\mathrm{dis},L}(i,j)=L_{\mathrm{dis}}(i,j+2)+2L_{\mathrm{dis}}(i+1,j+2)+L_{\mathrm{dis}}(i+2,j+2)\\ -L_{\mathrm{dis}}(i,j)-2L_{\mathrm{dis}}(i+1,j)-L_{\mathrm{dis}}(i+2,j)\end{array},$ Among them, L _org (i+2,j), L _org (i+2,j+1), L _org (i+2,j+2), L _org (i,j), L _org (i,j +1), L _org (i, j+2), L _org (i+1, j+2), and L _org (i+1, j) respectively indicate that the coordinate position in L _org is (i+2, j ), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1), (i,j+2), (i+1,j +2), (i+1,j) pixel value of the pixel, L _dis (i+2,j), L _dis (i+2,j+1), L _dis (i+2,j+2 ), L _dis (i,j), L _dis (i,j+1), L _dis (i,j+2), L _dis (i+1,j+2), L _dis (i+1,j ) respectively correspond to the coordinate positions in L _dis as (i+2,j), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1 ), (i,j+2), (i+1,j+2), (i+1,j) pixel values of the pixels.

The two images of R _org and R _dis are processed by horizontal and vertical Sobel operators respectively, and the corresponding horizontal gradient matrix maps and vertical gradient matrix maps of the two images of R _org and R _dis are respectively obtained, and the R _org The coefficient matrix of the corresponding gradient matrix map in the horizontal direction obtained after implementing the Sobel operator in the horizontal direction is denoted as I _h,org,R , for the coefficient value at the coordinate position (i,j) in I _h,org,R , denoted as I _h,org,R (i,j), $\begin{array}{c}{I}_{h,\mathrm{org},R}(i,j)=R_{\mathrm{org}}(i+2,j)+2R_{\mathrm{org}}(i+2,j+1)+R_{\mathrm{org}}(i+2,j+2)\\ -R_{\mathrm{org}}(i,j)-2R_{\mathrm{org}}(i,j+1)-R_{\mathrm{org}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding gradient matrix map in the vertical direction obtained after R _org implements the Sobel operator in the vertical direction is recorded as _Iv,org,R , and the coordinate position in _Iv,org,R is (i,j) The coefficient value of , denoted as I _v,org,R (i,j), $\begin{array}{c}{I}_{v,\mathrm{org},R}(i,j)=R_{\mathrm{org}}(i,j+2)+2R_{\mathrm{org}}(i+1,j+2)+R_{\mathrm{org}}(i+2,j+2)\\ -R_{\mathrm{org}}(i,j)-2R_{\mathrm{org}}(i+1,j)-R_{\mathrm{org}}(i+2,j)\end{array},$ The coefficient matrix of the corresponding horizontal gradient matrix map obtained after R _dis is processed by the horizontal direction Sobel operator is recorded as I _{h, dis, R} , and the coordinate position in I _{h, dis, R} is (i, j) The coefficient value of , denoted as I _h,dis,R (i,j), $\begin{array}{c}{I}_{h,\mathrm{dis},R}(i,j)=R_{\mathrm{dis}}(i+2,j)+2R_{\mathrm{dis}}(i+2,j+1)+R_{\mathrm{dis}}(i+2,j+2)\\ -R_{\mathrm{dis}}(i,j)-2R_{\mathrm{dis}}(i,j+1)-R_{\mathrm{dis}}(i,j+2)\end{array},$ The coefficient matrix of the corresponding gradient matrix map in the vertical direction obtained after R _dis is processed by the vertical direction Sobel operator is recorded as _Iv,dis,R , and the coordinate position in Iv _,dis,R is (i,j) The coefficient value of , denoted as I _v,dis,R (i,j), $\begin{array}{c}{I}_{v,\mathrm{dis},R}(i,j)=R_{\mathrm{dis}}(i,j+2)+2R_{\mathrm{dis}}(i+1,j+2)+R_{\mathrm{dis}}(i+2,j+2)\\ -R_{\mathrm{dis}}(i,j)-2R_{\mathrm{dis}}(i+1,j)-R_{\mathrm{dis}}(i+2,j)\end{array},$ Among them, R _org (i+2,j), R _org (i+2,j+1), R _org (i+2,j+2), R _org (i,j), R _org (i,j +1), R _org (i, j+2), R _org (i+1, j+2), and R _org (i+1, j) respectively indicate that the coordinate position in R _org is (i+2, j ), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1), (i,j+2), (i+1,j +2), (i+1,j) pixel value of the pixel, R _dis (i+2,j), R _dis (i+2,j+1), R _dis (i+2,j+2 ), R _dis (i,j), R _dis (i,j+1), R _dis (i,j+2), R _dis (i+1,j+2), R _dis (i+1,j ) respectively correspond to the coordinate positions in R _dis as (i+2,j), (i+2,j+1), (i+2,j+2), (i,j), (i,j+1 ), (i,j+2), (i+1,j+2), (i+1,j) pixel values of the pixels.

⑤ Calculate the structural direction distortion map of two overlapping blocks with the same coordinate position in the two images of L _org and L _dis , record the coefficient matrix of the structural direction distortion map as E _L , and the coordinate position in E _L is ( i,j), denote it as E _L (i,j), ${\mathrm{E.}}_L(i,j)=\frac{I_{h,\mathrm{org},L}(i,j)\×I_{h,\mathrm{dis},L}(i,j)+I_{v,\mathrm{org},L}(i,j)\×I_{v,\mathrm{dis},L}(i,j)+C_2}{\sqrt{{\left(I_{h,\mathrm{org},L}(i,j)\right)}^2+{\left(I_{v,\mathrm{org},L}(i,j)\right)}^2}\×\sqrt{{\left(I_{h,\mathrm{dis},L}(i,j)\right)}^2+{\left(I_{v,\mathrm{dis},L}(i,j)\right)}^2}+C_2},$ Among them, C ₂ represents a constant, and C ₂ is to avoid ${\mathrm{E.}}_L(i,j)=\frac{I_{h,\mathrm{org},L}(i,j)\×I_{h,\mathrm{dis},L}(i,j)+I_{v,\mathrm{org},L}(i,j)\×I_{v,\mathrm{dis},L}(i,j)+C_2}{\sqrt{{\left(I_{h,\mathrm{org},L}(i,j)\right)}^2+{\left(I_{v,\mathrm{org},L}(i,j)\right)}^2}\×\sqrt{{\left(I_{h,\mathrm{dis},L}(i,j)\right)}^2+{\left(I_{v,\mathrm{dis},L}(i,j)\right)}^2}+C_2}$ In the case where the denominator of is zero, C ₂ =0.02 may be chosen in the actual application process.

Calculate the structural direction distortion map of two overlapping blocks with the same coordinate position in the R _org and R _dis two images, record the coefficient matrix of the structural direction distortion map as E _R , and the coordinate position in E _R is (i ,j), which is recorded as E _R (i,j), ${\mathrm{E.}}_R(i,j)=\frac{I_{h,\mathrm{org},R}(i,j)\×I_{h,\mathrm{dis},R}(i,j)+I_{v,\mathrm{org},R}(i,j)\×I_{v,\mathrm{dis},R}(i,j)+C_2}{\sqrt{{\left(I_{h,\mathrm{org},R}(i,j)\right)}^2+{\left(I_{v,\mathrm{org},R}(i,j)\right)}^2}\×\sqrt{{\left(I_{h,\mathrm{dis},R}(i,j)\right)}^2+{\left(I_{v,\mathrm{dis},R}(i,j)\right)}^2}+C_2},$ C ₂ is to avoid ${\mathrm{E.}}_R(i,j)=\frac{I_{h,\mathrm{org},R}(i,j)\×I_{h,\mathrm{dis},R}(i,j)+I_{v,\mathrm{org},R}(i,j)\×I_{v,\mathrm{dis},R}(i,j)+C_2}{\sqrt{{\left(I_{h,\mathrm{org},R}(i,j)\right)}^2+{\left(I_{v,\mathrm{org},R}(i,j)\right)}^2}\×\sqrt{{\left(I_{h,\mathrm{dis},R}(i,j)\right)}^2+{\left(I_{v,\mathrm{dis},R}(i,j)\right)}^2}+C_2}$ In the case where the denominator of is zero, C ₂ =0.02 can be taken in the actual application process.

⑥ Calculate the structural distortion evaluation value of L _org and L _dis , denoted as Q ^L , Q ^L ＝ω ₁ ×Q _m,L +ω ₂ ×Q _nm,L , $Q_{m,L}=\frac{1}{N_{L,m}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_L(i,j)+{\mathrm{E.}}_L(i,j))\×A_L(i,j)),$ $N_{L,m}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}{A}_L(i,j),$ $Q_{\mathrm{nm},L}=\frac{1}{N_{L,\mathrm{nm}}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_L(i,j)+{\mathrm{E.}}_L(i,j))\×(1-A_L(i,j))),$ $N_{L,\mathrm{nm}}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(1-A_L(i,j)),$ Among them, ω ₁ represents the weight value of the sensitive area in L _org and L _dis , and ω ₂ represents the weight value of the non-sensitive area in L _org and L _dis .

Calculate the structural distortion evaluation value of R _org and R _dis , denoted as Q ^R , Q ^R ＝ω' ₁ ×Q _m,R +ω' ₂ ×Q _nm,R , $Q_{m,R}=\frac{1}{N_{R,m}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_R(i,j)+{\mathrm{E.}}_R(i,j))\×A_R(i,j)),$ $N_{R,m}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}{A}_R(i,j),$ $Q_{\mathrm{nm},R}=\frac{1}{N_{R,\mathrm{nm}}}\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(0.5\×(B_R(i,j)+{\mathrm{E.}}_R(i,j))\×(1-A_R(i,j))),$ $N_{R,\mathrm{nm}}=\underset{i=0}{\overset{W-8}{\mathrm{\Σ}}}\underset{j=0}{\overset{h-8}{\mathrm{\Σ}}}(1-A_R(i,j)),$ Among them, _ω'1 represents the weight value of the sensitive area in R _org and R _dis , and _ω'2 represents the weight value of the non-sensitive area in R _org and R _dis .

In this embodiment, FIG. 3 shows a block diagram for realizing the evaluation of image quality from the left viewpoint. Using 12 undistorted stereo images as shown in Figure 2a to Figure 2l, a distorted stereo image set consisting of 312 distorted stereo images is established, and the subjective quality of these 312 distorted stereo images is evaluated using a known subjective quality evaluation method Evaluation, the average subjective score difference (DMOS, Difference Mean Opinion Scores) of 312 distorted stereoscopic images, that is, the subjective quality score value of each distorted stereoscopic image. DMOS is the difference between the subjective score mean (MOS) and the full score (100), that is, DMOS=100-MOS, therefore, the larger the DMOS value, the worse the quality of the distorted stereoscopic image, and the smaller the DMOS value, the worse the quality of the distorted stereoscopic image. The better the quality, and the value range of DMOS is [0, 100]. On the other hand, for the above-mentioned 312 distorted stereoscopic images, calculate the corresponding Q _{m, L} and Q _{nm, L} of each distorted stereoscopic image according to steps ① to ⑥ of the method of the present invention; then use Q ^L = ω ₁ × Q _{m ,L} +(1-ω ₁ )×Q _nm,L is used for nonlinear fitting of the four-parameter Logistic function, and the values of α and ω ₁ are obtained. Here, three commonly used objective parameters for evaluating image quality evaluation methods are used as evaluation indicators, namely Pearson correlation coefficient (Correlation Coefficient, CC), Spearman correlation coefficient (SpearmanRank-Order Correlation Coefficient, SROCC) and mean square Error coefficient (Rooted Mean Squared Error, RMSE), CC reflects the accuracy of the objective model of the distorted stereo image evaluation function, and SROCC reflects the monotonicity between the objective model and subjective perception. RMSE reflects the accuracy of its prediction. The higher the value of CC and SROCC, the better the correlation between the stereoscopic image objective evaluation method and DMOS, and the lower the RMSE value, the better the correlation between the stereoscopic image objective evaluation method and DMOS. Figure 4a shows the CC performance change between the left view image quality and subjective perception quality of 312 stereo images under different α and ω _1, and Figure 4b shows the left view of 312 stereo images under different α and ω ₁ SROCC performance change between viewpoint image quality and subjective perception quality, Fig. 4c shows the RMSE performance variation between left viewpoint image quality and subjective perception quality of 312 stereo images under different α and ω ₁ , analysis Fig. 4a, From Figure 4b and Figure 4c, it can be seen that CC and SROCC will increase as the value of ω1 increases, while RMSE decreases as the value of _ω1 increases, indicating that the image quality of the left viewpoint is mainly determined by the quality of the sensitive area. However, the change of α value has little effect on the performance between left view image quality and subjective perception. Figure 5a shows the CC performance change between the left viewpoint image quality and subjective perception quality of 312 stereo images under different α in the case _of ω ₁ =1, ω ₂ =0; =1, ω ₂ =0, the SROCC performance change between the _left viewpoint image quality and subjective perception quality of 312 stereo images under different α; Fig. _5c shows the , the RMSE performance change between the left-viewpoint image quality and subjective perception quality of 312 stereo images under different α; analyzing Figure 5a, Figure 5b and Figure 5c, we can see that the CC, SROCC and RMSE values are all fluctuating on the percentile , but there is a peak. Therefore, in this embodiment, ω ₁ =1, α=2.1.

⑦ Calculate the spatial frequency similarity measure of the distorted stereo image S _dis to be evaluated relative to the original undistorted stereo image S _org according to Q ^L and Q ^R , denoted as Q _F , Q _F =β ₁ ×Q ^L +( 1-β ₁ )×Q ^R , where β ₁ represents the weight of Q ^L.

In this specific embodiment, the acquisition process of _β1 in step ⑦ is:

⑦-1. Using n undistorted stereoscopic images to establish a distorted stereoscopic image set under different distortion types and different degrees of distortion, the distorted stereoscopic image set includes multiple distorted stereoscopic images, wherein n≥1.

⑦-2. Utilize the subjective quality evaluation method to obtain the average subjective score difference value of each distorted stereoscopic image in the distorted stereoscopic image set, which is denoted as DMOS, DMOS=100-MOS, wherein, MOS represents the mean value of the subjective score, and DMOS∈[0,100 ].

⑦-3, according to the operation process of step ① to step ⑥, calculate the evaluation value Q of the sensitive area of the left viewpoint image of each distorted stereoscopic image in the distorted stereoscopic image set relative to the left viewpoint image of the corresponding undistorted stereoscopic image _m,L and the evaluation value Q _nm,L of the non-sensitive area.

⑦-4. Using a mathematical fitting method to fit the average subjective score difference DMOS and the corresponding Q _m,L and Q _nm,L of the distorted stereo images in the distorted stereo image set, so as to obtain the value of β ₁ .

In this embodiment, _β1 determines the contribution of Q ^L to the stereoscopic image quality. For blockiness, the stereoscopic image quality is about half of the sum of the quality of the left-viewpoint image and the right-viewpoint image. For blurring and distortion, the stereoscopic image The quality is mainly determined by the viewpoint with better quality. Because the left-viewpoint image and right-viewpoint image of the stereoscopic image test library are distorted to the same degree at the same time, the quality of the left-viewpoint image and the quality of the right-viewpoint image have little change, so the change of _β1 has little effect on the subjective performance of the stereoscopic image. First, the above-mentioned 312 distorted stereoscopic images are calculated according to steps ① to ⑥ of the method of the present invention to obtain the corresponding ^QL and ^QR of each distorted stereoscopic image, and then the four-parameter fitting is used to obtain the _β1 value. Figure 6a shows the CC performance change between the quality of the left and right viewpoint images and the subjective perception quality under different β ₁ , and Fig. 6a shows the SROCC performance between the quality of the left and right viewpoint images and the subjective perception quality under different β ₁ Figure 6a shows the RMSE performance change between the quality of the left and right viewpoint images and the subjective perception quality under different β _1. Analyzing Figure 6a, Figure 6b and Figure 6c, we can know that with the change of β ₁ value, CC, SROCC and RMSE values have little change, fluctuating on the percentile, but both have peaks. Here, β ₁ =0.5 is taken.

⑧ Calculate the absolute difference image matrix of L _org and R _org means that Calculate the absolute difference image matrix of L _dis and R _dis means that Among them, "||" is the absolute value symbol.

⑨ will and The image is divided into non-overlapping blocks of size 8×8, and then and Singular value decomposition is performed on all the blocks in the image respectively to get The corresponding singular value map composed of the singular value matrix of each block and The corresponding singular value map composed of the singular value matrix of each block will be The coefficient matrix of the singular value map obtained after implementing singular value decomposition is denoted as G _org , and for the singular value at the coordinate position (p, q) in the singular value matrix of the nth block in G _org , it is denoted as Will The coefficient matrix of the singular value map obtained after the implementation of singular value decomposition is denoted as G _dis , and for the singular value at the coordinate position (p, q) in the singular value matrix of the nth block in G _dis , it is denoted as Among them, W _LR means and The width, H _LR means and height of, 0≤p≤7,0≤q≤7.

Here, in order to reduce computational complexity, an 8×8 block does not have repeated columns or rows with its most adjacent left or right block or upper or lower block, that is, blocks do not overlap each other.

⑩ Calculation The corresponding singular value maps and The singular value deviation evaluation value of the corresponding singular value map, denoted as K, $K=\frac{64}{W_{\mathrm{LR}}\×h_{\mathrm{LR}}}\×\underset{\mathrm{no}=0}{\overset{W_{\mathrm{LR}}\×h_{\mathrm{LR}}/64-1}{\mathrm{\Σ}}}\frac{\underset{p=0}{\overset{7}{\mathrm{\Σ}}}(G_{\mathrm{org}}^{\mathrm{no}}(p,p)\×|G_{\mathrm{org}}^{\mathrm{no}}(p,p)-G_{\mathrm{dis}}^{\mathrm{no}}(p,p)\left|\right)}{\underset{p=0}{\overset{7}{\mathrm{\Σ}}}{G}_{\mathrm{org}}^{\mathrm{no}}(p,p)},$ in, Indicates the singular value at the coordinate position (p,p) in the singular value matrix of the nth block in G _org , Indicates the singular value at the coordinate position (p, p) in the singular value matrix of the nth block in _Gdis .

right and Singular value decomposition is carried out respectively, and we get and Corresponding to two orthogonal matrices and one singular value matrix, the The two orthogonal matrices obtained after implementing singular value decomposition are denoted as χ _org and V _org respectively, and the The singular value matrix obtained after implementing singular value decomposition is denoted as O _org , Will The two orthogonal matrices obtained after implementing singular value decomposition are denoted as χ _dis and V _dis respectively, and the The singular value matrix obtained after implementing singular value decomposition is denoted as O _dis ,

Calculate separately and The residual matrix image after depriving the singular value of the image, the The residual matrix after deprivation of singular values is denoted as X _org , X _org = χ _org × Λ × V _org , the The residual matrix after stripping the singular value is marked as X _dis , X _dis =χ _dis ×Λ×V _dis , where Λ represents the identity matrix, and the size of Λ is consistent with the size of O _org and O _dis .

Calculate the mean deviation rate of X _org and X _dis , denoted as Wherein, x represents the abscissa of the pixel in X _org and X _dis , and y represents the ordinate of the pixel in X _org and X _dis .

Calculate the stereo perception evaluation metric of the distorted stereo image S _dis to be evaluated relative to the original undistorted stereo image S _org , denoted as Q _S , where τ represents a constant used to adjust K and The importance played in _QS .

In the present embodiment, first obtain the absolute difference images of the above-mentioned 312 distorted stereo images and 10 undistorted stereo images, and then follow steps 8 to 8 of the method of the present invention. Calculate the K and K corresponding to each distorted stereo image Here, the value of τ determines the importance of singular value deviation and residual information in depth perception evaluation. Figure 7a shows the CC performance change between the stereo perception quality and subjective perception of 312 distorted stereo images under different τ, and Figure 7b shows the stereo perception quality and subjective perception of 312 distorted stereo images under different τ The variation of SROCC performance between perceptions, Fig. 7c shows the variation of RMSE performance between the stereo perception quality and subjective perception of 312 distorted stereo images under different τ, in Fig. 7a, Fig. 7b and Fig. 7c, τ is in [- 164], analyzing Figure 7a, Figure 7b and Figure 7c, it can be seen that there is an extreme value in CC, SROCC and RMSE, and the change of τ, and the position is roughly the same, here we take τ=-8.

According to Q _F and Q _S , calculate the image quality evaluation score of the distorted stereo image S _dis to be evaluated, denoted as Q, Q=Q _F ×(Q _S ) ^ρ , where ρ represents the weight coefficient value.

In the present embodiment, for the above-mentioned 312 distorted stereoscopic images according to the method steps of the present invention from ① to Calculate the corresponding Q _F and Q _S of each distorted stereoscopic image, and then use Q=Q _F ×(Q _S ) ^ρ to perform nonlinear fitting of the four-parameter Logistic function to obtain ρ, and the value of ρ determines the left and right viewpoint images Quality and Stereoscopic Quality Contributions in Stereoscopic Image Quality. Both the Q _F and Q _S values become smaller as the degree of distortion of the stereoscopic image deepens, so the value range of the ρ value is greater than 0. Figure 8a shows the CC performance variation between the quality of 312 stereo images and subjective perceptual quality under different ρ values, and Figure 8b shows the relationship between the quality of 312 stereo images and subjective perceptual quality under different ρ values. The change of SROCC performance among them. Figure 8c shows the RMSE performance change between the quality of 312 stereoscopic images and the subjective perception quality under different ρ values. Analyzing Figure 8a, Figure 8b, and Figure 8c, it can be known that the value of ρ is too large or too small will affect the consistency between the stereoscopic image quality objective evaluation model and subjective perception. With the change of ρ value, there are extreme points in the CC, SROCC and RMSE values, and the approximate positions are the same. Here, ρ=0.3 .

The correlation between the final evaluation result of the image quality evaluation function Q=Q _F ×(Q _S ) ^0.3 of the distorted stereoscopic image obtained in this embodiment and the subjective score DMOS is analyzed. First, the output value Q of the final stereoscopic image quality evaluation result calculated by the image quality evaluation function Q=Q _F * (Q _S ) ^0.3 of the distorted stereoscopic image obtained in this embodiment, then the output value Q is done as a four-parameter Logistic function Non-linear fitting, and finally obtain the performance index value between the stereoscopic objective evaluation model and the subjective perception. Here, four commonly used objective parameters for evaluating image quality evaluation methods are used as evaluation indicators, namely CC, SROCC, Outlier Ratio (OR), and RMSE. OR reflects the degree of dispersion of the stereoscopic image quality objective rating model, that is, the proportion of the number of distorted stereoscopic images whose difference between the evaluation value after four-parameter fitting and DMOS is greater than a certain threshold in all distorted stereoscopic images. Table 1 provides the evaluation performance CC, SROCC, OR and RMSE coefficients, as can be seen from the data in Table 1, the image quality evaluation function Q=Q _F * (Q _S ) ^0.3 of the distorted stereoscopic image obtained by the present embodiment is obtained by calculating The correlation between the output value Q of the final evaluation result and the subjective score DMOS is very high, the CC value and SROCC value both exceed 0.92, and the RMSE value is lower than 6.5, indicating that the objective evaluation result is more consistent with the subjective perception of the human eye. The effectiveness of the method of the present invention is illustrated.

Table 1 Correlation between the image quality evaluation score and the subjective score of the distorted stereoscopic image obtained in this implementation

Gblur JP2K JPEG WN H264 ALL number 60 60 60 60 72 312 CC 0.9658 0.9479 0.9533 0.9554 0.9767 0.9235 SROCC 0.9655 0.9489 0.9524 0.9274 0.9545 0.9430 OR 0 0 0 0 0 0 RMSE 5.4719 3.8180 4.3010 4.6151 3.0135 6.5890

Claims (3)

1. A three-dimensional image objective quality evaluation method based on structural distortion is characterized by comprising the following steps:

making S_orgFor original undistorted stereo image, let S_disFor the distorted stereo image to be evaluated, the original undistorted stereo image S is taken_orgIs recorded as L_orgThe original undistorted stereo image S is processed_orgIs recorded as R_orgDistorted stereoscopic image S to be evaluated_disIs recorded as L_disDistorted stereoscopic image S to be evaluated_disIs recorded as R_dis；

② to L_orgAnd L_dis、R_orgAnd R_disRespectively dividing the 4 images into regions to respectively obtain L_orgAnd L_dis、R_orgAnd R_disMapping L sensitive area matrixes corresponding to the 4 images_orgAnd L_disRespectively carrying out region division to obtain coefficient matrixes of respectively corresponding sensitive region matrix mapping graphs, and recording the coefficient matrixes as A_LFor A_LThe value of the coefficient at the middle coordinate position (i, j), which is denoted as A_L(i,j)，R is to be_orgAnd R_disRespectively carrying out region division to obtain coefficient matrixes of respectively corresponding sensitive region matrix mapping graphs, and recording the coefficient matrixes as A_RFor A_RThe value of the coefficient at the middle coordinate position (i, j), which is denoted as A_R(i,j)，Wherein, i is not less than 0 and not more than (W-8), j is not less than 0 and not more than (H-8), W represents L_org、L_dis、R_orgAnd R_disH represents L_org、L_dis、R_orgAnd R_disHigh of (d);

③ will L_orgAnd L_disThe 2 images were divided into (W-7) × (H-7) overlapping blocks of size 8 × 8, respectively, and L was calculated_orgAnd L_disThe structural amplitude distortion mapping maps of two overlapped blocks with the same coordinate positions in 2 images are recorded as B in the coefficient matrix of the structural amplitude distortion mapping maps_LFor B_LThe value of the coefficient at the middle coordinate position (i, j), which is denoted as B_L(i,j)， <math> <mrow> <msub> <mi>B</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mn>2</mn> <mo>&times;</mo> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> </mrow> <mrow> <msup> <mrow> <mo>(</mo> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>&sigma;</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> </mrow> </mfrac> <mo>,</mo> </mrow> </math> Wherein, B_L(i, j) also represents L_orgThe upper left corner coordinate position is (i, j) and L_disThe middle upper left corner coordinate position is the structural amplitude distortion value of the overlapped block with the size of (i, j) being 8 multiplied by 8, <math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msqrt> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>L</mi> <mi>org</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>U</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msub> <mi>L</mi> <mi>org</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msqrt> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>L</mi> <mi>dis</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>U</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msub> <mi>L</mi> <mi>dis</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <mrow> <mo>(</mo> <mrow> <mo>(</mo> <msub> <mi>L</mi> <mi>org</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>&times;</mo> <mrow> <mo>(</mo> <msub> <mi>L</mi> <mi>dis</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> L_org(i + x, j + y) represents L_orgThe pixel value L of the pixel point with the middle coordinate position of (i + x, j + y)_dis(i + x, j + y) represents L_disThe pixel value C of the pixel point with the middle coordinate position of (i + x, j + y)₁Represents a constant, where 0. ltoreq. i.ltoreq.W-8, 0. ltoreq. j.ltoreq.H-8;

r is to be_orgAnd R_disThe 2 images were divided into (W-7) × (H-7) overlapping blocks of size 8 × 8, respectively, and R was calculated_orgAnd R_disThe structural amplitude distortion mapping maps of two overlapped blocks with the same coordinate positions in 2 images are recorded as B in the coefficient matrix of the structural amplitude distortion mapping maps_RFor B_RThe value of the coefficient at the middle coordinate position (i, j), which is denoted as B_R(i,j)， <math> <mrow> <msub> <mi>B</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mn>2</mn> <mo>&times;</mo> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> </mrow> <mrow> <msup> <mrow> <mo>(</mo> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>&sigma;</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> </mrow> </mfrac> <mo>,</mo> </mrow> </math> Wherein, B_R(i, j) also represents R_orgThe size of the overlapped block with the upper left corner coordinate position of (i, j) being 8 multiplied by 8 and R_disThe middle upper left corner coordinate position is the structural amplitude distortion value of the overlapped block with the size of (i, j) being 8 multiplied by 8, <math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msqrt> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>R</mi> <mi>org</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>U</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msub> <mi>R</mi> <mi>org</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msqrt> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>R</mi> <mi>dis</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>U</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msub> <mi>R</mi> <mi>dis</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>64</mn> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <mrow> <mo>(</mo> <mrow> <mo>(</mo> <msub> <mi>R</mi> <mi>org</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>&times;</mo> <mrow> <mo>(</mo> <msub> <mi>R</mi> <mi>dis</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>+</mo> <mi>x</mi> <mo>,</mo> <mi>j</mi> <mo>+</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>U</mi> <mrow> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> R_org(i + x, j + y) represents R_orgThe pixel value R of the pixel point with the middle coordinate position of (i + x, j + y)_dis(i + x, j + y) represents R_disThe pixel value C of the pixel point with the middle coordinate position of (i + x, j + y)₁Represents a constant, where 0. ltoreq. i.ltoreq.W-8, 0. ltoreq. j.ltoreq.H-8;

fourthly, to L_orgAnd L_disRespectively carrying out Sobel operator processing in the horizontal direction and the vertical direction on the 2 images to respectively obtain L_orgAnd L_disThe gradient matrix mapping chart in the horizontal direction and the gradient matrix mapping chart in the vertical direction corresponding to the 2 images respectively are converted into L_orgThe coefficient matrix of the corresponding horizontal gradient matrix mapping graph obtained after horizontal Sobel operator processing is carried out is marked as I_h,org,LFor I_h,org,LThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_h,org,L(i,j)， $\begin{array}{c}{I}_{h,\mathrm{org},L}(i,j)=L_{\mathrm{org}}(i+2,j)+2L_{\mathrm{org}}(i+2,j+1)+L_{\mathrm{org}}(i+2,j+2)\\ -L_{\mathrm{org}}(i,j)-2L_{\mathrm{org}}(i,j+1)-L_{\mathrm{org}}(i,j+2)\end{array},$ Mixing L with_orgThe coefficient matrix of the corresponding vertical gradient matrix mapping graph obtained after the vertical Sobel operator is implemented is marked as I_v,org,LFor I_v,org,LThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_v,org,L(i,j)， $\begin{array}{c}{I}_{v,\mathrm{org},L}(i,j)=L_{\mathrm{org}}(i,j+2)+2L_{\mathrm{org}}(i+1,j+2)+L_{\mathrm{org}}(i+2,j+2)\\ -L_{\mathrm{org}}(i,j)-2L_{\mathrm{org}}(i+1,j)-L_{\mathrm{org}}(i+2,j)\end{array},$ Mixing L with_disThe coefficient matrix of the corresponding horizontal gradient matrix mapping graph obtained after horizontal Sobel operator processing is carried out is marked as I_h,dis,LFor I_h,dis,LThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_h,dis,L(i,j)， $\begin{array}{c}{I}_{h,\mathrm{dis},L}(i,j)=L_{\mathrm{dis}}(i+2,j)+2L_{\mathrm{dis}}(i+2,j+1)+L_{\mathrm{dis}}(i+2,j+2)\\ -L_{\mathrm{dis}}(i,j)-2L_{\mathrm{dis}}(i,j+1)-L_{\mathrm{dis}}(i,j+2)\end{array},$ Mixing L with_disThe coefficient matrix of the corresponding vertical gradient matrix mapping graph obtained after the vertical Sobel operator is implemented is marked as I_v,dis,LFor I_v,dis,LThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_v,dis,L(i,j)， $\begin{array}{c}{I}_{v,\mathrm{dis},L}(i,j)=L_{\mathrm{dis}}(i,j+2)+2L_{\mathrm{dis}}(i+1,j+2)+L_{\mathrm{dis}}(i+2,j+2)\\ -L_{\mathrm{dis}}(i,j)-2L_{\mathrm{dis}}(i+1,j)-L_{\mathrm{dis}}(i+2,j)\end{array},$ Wherein L is_org(i+2,j)、L_org(i+2,j+1)、L_org(i+2,j+2)、L_org(i,j)、L_org(i,j+1)、L_org(i,j+2)、L_org(i+1,j+2)、L_org(i +1, j) each represents L_orgThe pixel values, L, of the pixel points having the middle coordinate positions of (i +2, j), (i +2, j +1), (i +2, j +2), (i, j +1), (i, j +2), (i +1, j +2), and (i +1, j)_dis(i+2,j)、L_dis(i+2,j+1)、L_dis(i+2,j+2)、L_dis(i,j)、L_dis(i,j+1)、L_dis(i,j+2)、L_dis(i+1,j+2)、L_dis(i +1, j) correspond to each otherRepresents L_disThe middle coordinate position is the pixel value of the pixel point of (i +2, j), (i +2, j +1), (i +2, j +2), (i, j +1), (i, j +2), (i +1, j +2), or (i +1, j);

to R_orgAnd R_disRespectively carrying out Sobel operator processing in the horizontal direction and the vertical direction on the 2 images to respectively obtain R_orgAnd R_disThe gradient matrix mapping map in the horizontal direction and the gradient matrix mapping map in the vertical direction corresponding to the 2 images are respectively obtained by mapping R_orgThe coefficient matrix of the corresponding horizontal gradient matrix mapping graph obtained after horizontal Sobel operator processing is carried out is marked as I_h,org,RFor I_h,org,RThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_h,org,R(i,j)， $\begin{array}{c}{I}_{h,\mathrm{org},R}(i,j)=R_{\mathrm{org}}(i+2,j)+2R_{\mathrm{org}}(i+2,j+1)+R_{\mathrm{org}}(i+2,j+2)\\ -R_{\mathrm{org}}(i,j)-2R_{\mathrm{org}}(i,j+1)-R_{\mathrm{org}}(i,j+2)\end{array},$ R is to be_orgThe coefficient matrix of the corresponding vertical gradient matrix mapping graph obtained after the vertical Sobel operator is implemented is marked as I_v,org,RFor I_v,org,RThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_v,org,R(i,j)， $\begin{array}{c}{I}_{v,\mathrm{org},R}(i,j)=R_{\mathrm{org}}(i,j+2)+2R_{\mathrm{org}}(i+1,j+2)+R_{\mathrm{org}}(i+2,j+2)\\ -R_{\mathrm{org}}(i,j)-2R_{\mathrm{org}}(i+1,j)-R_{\mathrm{org}}(i+2,j)\end{array},$ R is to be_disThe coefficient matrix of the corresponding horizontal gradient matrix mapping graph obtained after horizontal Sobel operator processing is carried out is marked as I_h,dis,RFor I_h,dis,RThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_h,dis,R(i,j)， $\begin{array}{c}{I}_{h,\mathrm{dis},R}(i,j)=R_{\mathrm{dis}}(i+2,j)+2R_{\mathrm{dis}}(i+2,j+1)+R_{\mathrm{dis}}(i+2,j+2)\\ -R_{\mathrm{dis}}(i,j)-2R_{\mathrm{dis}}(i,j+1)-R_{\mathrm{dis}}(i,j+2)\end{array},$ R is to be_disThe coefficient matrix of the corresponding vertical gradient matrix mapping graph obtained after the vertical Sobel operator is implemented is marked as I_v,dis,RFor I_v,dis,RThe value of the coefficient at the middle coordinate position (I, j), which is denoted as I_v,dis,R(i,j)， $\begin{array}{c}{I}_{v,\mathrm{dis},R}(i,j)=R_{\mathrm{dis}}(i,j+2)+2R_{\mathrm{dis}}(i+1,j+2)+R_{\mathrm{dis}}(i+2,j+2)\\ -R_{\mathrm{dis}}(i,j)-2R_{\mathrm{dis}}(i+1,j)-R_{\mathrm{dis}}(i+2,j)\end{array},$ Wherein R is_org(i+2,j)、R_org(i+2,j+1)、R_org(i+2,j+2)、R_org(i,j)、R_org(i,j+1)、R_org(i,j+2)、R_org(i+1,j+2)、R_org(i +1, j) each represents R_orgThe middle coordinate position is the pixel value, R, of the pixel points of (i +2, j), (i +2, j +1), (i +2, j +2), (i, j +1), (i, j +2), (i +1, j +2), and (i +1, j)_dis(i+2,j)、R_dis(i+2,j+1)、R_dis(i+2,j+2)、R_dis(i,j)、R_dis(i,j+1)、R_dis(i,j+2)、R_dis(i+1,j+2)、R_dis(i +1, j) each represents R_disThe middle coordinate position is the pixel value of the pixel point of (i +2, j), (i +2, j +1), (i +2, j +2), (i, j +1), (i, j +2), (i +1, j +2), or (i +1, j);

calculating L_orgAnd L_disThe structural direction distortion mapping maps of two overlapped blocks with the same coordinate positions in 2 images are recorded as E in the coefficient matrix of the structural direction distortion mapping maps_LFor E_LThe value of the coefficient at the middle coordinate position (i, j), denoted as E_L(i,j)， <math> <mrow> <msub> <mi>E</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> </mrow> <mrow> <msqrt> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>&times;</mo> <msqrt> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>+</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> </mrow> </mfrac> <mo>,</mo> </mrow> </math> Wherein, C₂Represents a constant;

calculation of R_orgAnd R_disThe structural direction distortion mapping maps of two overlapped blocks with the same coordinate positions in 2 images are recorded as E in the coefficient matrix of the structural direction distortion mapping maps_RFor E_RThe middle coordinate position is the coefficient value at (i, j), which is denoted as ER (i, j), <math> <mrow> <msub> <mi>E</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> </mrow> <mrow> <msqrt> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>org</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>&times;</mo> <msqrt> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>h</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mrow> <mi>v</mi> <mo>,</mo> <mi>dis</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </msqrt> <mo>+</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> </mrow> </mfrac> <mo>;</mo> </mrow> </math>

sixthly, calculating L_orgAnd L_disStructural distortion evaluation value of (1), denoted as Q^L，Q^L＝ω₁×Q_m,L+ω₂×Q_nm,L， <math> <mrow> <msub> <mi>Q</mi> <mrow> <mi>m</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>L</mi> <mo>,</mo> <mi>m</mi> </mrow> </msub> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <mrow> <mo>(</mo> <mn>0.5</mn> <mo>&times;</mo> <mrow> <mo>(</mo> <msub> <mi>B</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>E</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>A</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>N</mi> <mrow> <mi>L</mi> <mo>,</mo> <mi>m</mi> </mrow> </msub> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <msub> <mi>A</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>Q</mi> <mrow> <mi>nm</mi> <mo>,</mo> <mi>L</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>L</mi> <mo>,</mo> <mi>nm</mi> </mrow> </msub> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <mrow> <mo>(</mo> <mn>0.5</mn> <mo>&times;</mo> <mrow> <mo>(</mo> <msub> <mi>B</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>E</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>&times;</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>A</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>N</mi> <mrow> <mi>L</mi> <mo>,</mo> <mi>nm</mi> </mrow> </msub> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>A</mi> <mi>L</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> Wherein, ω is₁Represents L_orgAnd L_disWeight value of middle sensitive area, omega₂Represents L_orgAnd L_disThe weight value of the middle non-sensitive area;

calculation of R_orgAnd R_disStructural distortion evaluation value of (1), denoted as Q^R，Q^R＝ω'₁×Q_m,R+ω'₂×Q_nm,R， <math> <mrow> <msub> <mi>Q</mi> <mrow> <mi>m</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>R</mi> <mo>,</mo> <mi>m</mi> </mrow> </msub> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <mrow> <mo>(</mo> <mn>0.5</mn> <mo>&times;</mo> <mrow> <mo>(</mo> <msub> <mi>B</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>E</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>&times;</mo> <msub> <mi>A</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>N</mi> <mrow> <mi>R</mi> <mo>,</mo> <mi>m</mi> </mrow> </msub> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <msub> <mi>A</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>Q</mi> <mrow> <mi>nm</mi> <mo>,</mo> <mi>R</mi> </mrow> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>R</mi> <mo>,</mo> <mi>nm</mi> </mrow> </msub> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <mrow> <mo>(</mo> <mn>0.5</mn> <mo>&times;</mo> <mrow> <mo>(</mo> <msub> <mi>B</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>E</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>&times;</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>A</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> <math> <mrow> <msub> <mi>N</mi> <mrow> <mi>R</mi> <mo>,</mo> <mi>nm</mi> </mrow> </msub> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>W</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <mi>H</mi> <mo>-</mo> <mn>8</mn> </mrow> </munderover> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>A</mi> <mi>R</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> Wherein, omega'₁Represents R_orgAnd R_disWeight value of Medium sensitive region, ω'₂Represents R_orgAnd R_disThe weight value of the middle non-sensitive area;

is according to Q^LAnd Q^RCalculating a distorted stereo image S to be evaluated_disRelative to the original undistorted stereo image S_orgThe spatial frequency similarity measure of (2), denoted as Q_F，Q_F＝β₁×Q^L+(1-β₁)×Q^RWherein, β₁Represents Q^LThe weight of (2);

b calculation of L_orgAnd R_orgIs determined by the absolute difference image of (a),expressed in matrix form as Calculating L_disAnd R_disIs expressed in matrix form as Wherein, "|" is an absolute value symbol;

ninthly willAndthe images being divided intoBlocks of size 8 x 8 which do not overlap with each other, and thenAndrespectively performing singular value decomposition on all blocks in the images to obtainCorresponding singular value map composed of singular value matrix of each block thereof andcorresponding singular value matrix of each blockA map of singular values of the compositionThe coefficient matrix of the singular value mapping graph obtained after singular value decomposition is recorded as G_orgFor G_orgThe coordinate position in the singular value matrix of the nth block is the singular value at (p, q), which is recorded asWill be provided withThe coefficient matrix of the singular value mapping graph obtained after singular value decomposition is recorded as G_disFor G_disThe coordinate position in the singular value matrix of the nth block is the singular value at (p, q), which is recorded asWherein, W_LRTo representAndwidth of (H)_LRTo representAndis high in the direction of the horizontal axis,0≤p≤7,0≤q≤7；

calculation of rCorresponding singular value map andthe singular value deviation evaluation value of the corresponding singular value map, denoted as K, <math> <mrow> <mi>K</mi> <mo>=</mo> <mfrac> <mn>64</mn> <mrow> <msub> <mi>W</mi> <mi>LR</mi> </msub> <mo>&times;</mo> <msub> <mi>H</mi> <mi>LR</mi> </msub> </mrow> </mfrac> <mo>&times;</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>n</mi> <mo>=</mo> <mn>0</mn> </mrow> <mrow> <msub> <mi>W</mi> <mi>LR</mi> </msub> <mo>&times;</mo> <msub> <mi>H</mi> <mi>LR</mi> </msub> <mo>/</mo> <mn>64</mn> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <mfrac> <mrow> <munderover> <mi>&Sigma;</mi> <mrow> <mi>p</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <mrow> <mo>(</mo> <msubsup> <mi>G</mi> <mi>org</mi> <mi>n</mi> </msubsup> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>&times;</mo> <mo>|</mo> <msubsup> <mi>G</mi> <mi>org</mi> <mi>n</mi> </msubsup> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>-</mo> <msubsup> <mi>G</mi> <mi>dis</mi> <mi>n</mi> </msubsup> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>|</mo> <mo>)</mo> </mrow> </mrow> <mrow> <munderover> <mi>&Sigma;</mi> <mrow> <mi>p</mi> <mo>=</mo> <mn>0</mn> </mrow> <mn>7</mn> </munderover> <msubsup> <mi>G</mi> <mi>org</mi> <mi>n</mi> </msubsup> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>p</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>,</mo> </mrow> </math> wherein,represents G_orgThe coordinate position in the singular value matrix of the nth block is the singular value at (p, p),represents G_disThe coordinate position in the singular value matrix of the nth block is a singular value at (p, p);

to pairAndrespectively carrying out singular value decomposition to respectively obtainAnd2 orthogonal matrixes and 1 singular value matrix corresponding to each other are used for generating a plurality of singular value matrixes2 orthogonal matrixes obtained after singular value decomposition are respectively recorded as chi_orgAnd V_orgWill beThe singular value matrix obtained after singular value decomposition is recorded as O_org，Will be provided with2 orthogonal matrixes obtained after singular value decomposition are respectively recorded as chi_disAnd V_disWill beThe singular value matrix obtained after singular value decomposition is recorded as O_dis，

Respectively calculateAndresidual matrix image of the image after the singular value is deprivedThe residual matrix image after the singular value deprivation is marked as X_org，X_org＝χ_org×Λ×V_orgWill beThe residual matrix image after the singular value deprivation is marked as X_dis，X_dis＝χ_dis×Λ×V_disWherein Λ represents the identity matrix, the size of Λ and O_orgAnd O_disAre consistent in size;

calculating X_orgAnd X_disThe mean deviation ratio of (D) is recorded as Wherein X represents X_orgAnd X_disThe abscissa of the pixel point in (1) and y represents X_orgAnd X_disThe ordinate of the pixel point in (1);

calculating a distorted stereo image S to be evaluated_disRelative to the original undistorted stereo image S_orgThe stereo perception evaluation metric of (1), denoted as Q_S，Wherein τ represents a constant for adjusting K andat Q_SThe importance of (1);

according to Q_FAnd Q_STo calculateDistorted stereoscopic image S of evaluation_disThe image quality evaluation score of (1) is Q, Q is Q_F×(Q_S)^ρWhere ρ represents a weight coefficient value.

2. The method for evaluating the objective quality of a stereoscopic image based on structural distortion according to claim 1, wherein L is_orgAnd L_disCoefficient matrix A of the respective corresponding sensitive area matrix map_LThe acquisition process comprises the following steps:

② -a1, pair L_orgPerforming Sobel operator processing in horizontal and vertical directions to obtain L_orgRespectively, as Z, in the horizontal direction and in the vertical direction_h,l1And Z_v,l1Then calculate L_orgIs marked as Z_l1，Wherein Z is_l1(x, y) represents Z_l1Gradient amplitude, Z, of pixel point with (x, y) as middle coordinate position_h,l1(x, y) represents Z_h,l1The horizontal gradient value Z of the pixel point with the (x, y) middle coordinate position_v,l1(x, y) represents Z_v,l1The gradient value in the vertical direction of the pixel point with the middle coordinate position (x, y) is that x is more than or equal to 1 and less than or equal to W ', y is more than or equal to 1 and less than or equal to H ', wherein W ' represents Z_l1Width of (A), H' represents Z_l1High of (d);

② -a2, pair L_disPerforming Sobel operator processing in horizontal and vertical directions to obtain L_disRespectively, as Z, in the horizontal direction and in the vertical direction_h,l2And Z_v,l2Then calculate L_disIs marked as Z_l2，Wherein Z is_l2(x, y) represents Z_l2Gradient amplitude, Z, of pixel point with (x, y) as middle coordinate position_h,l2(x, y) represents Z_h,l2The horizontal gradient value Z of the pixel point with the (x, y) middle coordinate position_v,l2(x, y) represents Z_v,l2The gradient value in the vertical direction of the pixel point with the middle coordinate position (x, y) is that x is more than or equal to 1 and less than or equal to W ', y is more than or equal to 1 and less than or equal to H ', wherein W ' represents Z_l2Width of (A), H' represents Z_l2High of (d);

a3, calculating the threshold T needed when dividing the area, <math> <mrow> <mi>T</mi> <mo>=</mo> <mi>&alpha;</mi> <mo>&times;</mo> <mfrac> <mn>1</mn> <mrow> <msup> <mi>W</mi> <mo>&prime;</mo> </msup> <mo>&times;</mo> <msup> <mi>H</mi> <mo>&prime;</mo> </msup> </mrow> </mfrac> <mo>&times;</mo> <mrow> <mo>(</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>W</mi> <mo>&prime;</mo> </msup> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>H</mi> <mo>&prime;</mo> </msup> </munderover> <msub> <mi>Z</mi> <mrow> <mi>l</mi> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>+</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>W</mi> <mo>&prime;</mo> </msup> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>H</mi> <mo>&prime;</mo> </msup> </munderover> <msub> <mi>Z</mi> <mrow> <mi>l</mi> <mn>2</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> wherein α is a constant, Z_l1(x, y) represents Z_l1Gradient amplitude, Z, of pixel point with (x, y) as middle coordinate position_l2(x, y) represents Z_l2The gradient amplitude of the pixel point with the middle coordinate position (x, y);

② a4, mixing Z_l1The gradient amplitude of the pixel point with the middle coordinate position (i, j) is recorded as Z_l1(i, j) reacting Z_l2The gradient amplitude of the pixel point with the middle coordinate position (i, j) is recorded as Z_l2(i, j), judgment of Z_l1(i,j)>T or Z_l2(i,j)>Whether T is true, if so, then L is determined_orgAnd L_disThe pixel point with the middle coordinate position (i, j) belongs to the sensitive area, and order A_L(i, j) ═ 1, otherwise, determine L_orgAnd L_disThe pixel point with the middle coordinate position (i, j) belongs to the non-sensitive area, and order A_L(i, j) is 0, wherein i is more than or equal to 0 and less than or equal to (W-8), and j is more than or equal to 0 and less than or equal to (H-8);

in the step II, R_orgAnd R_disCoefficient matrix A of the respective corresponding sensitive area matrix map_RThe acquisition process comprises the following steps:

② -b1, vs R_orgPerforming Sobel operator processing in horizontal and vertical directions to obtain R_orgRespectively, as Z, in the horizontal direction and in the vertical direction_h,r1And Z_v,r1Then calculating R_orgIs marked as Z_r1，Wherein Z is_r1(x, y) represents Z_r1Gradient amplitude, Z, of pixel point with (x, y) as middle coordinate position_h,r1(x, y) represents Z_h,r1The horizontal gradient value Z of the pixel point with the (x, y) middle coordinate position_v,r1(x, y) represents Z_v,r1The gradient value in the vertical direction of the pixel point with the middle coordinate position (x, y) is that x is more than or equal to 1 and less than or equal to W ', y is more than or equal to 1 and less than or equal to H ', wherein W ' represents Z_r1Is wideH' represents Z_r1High of (d);

② -b2, vs R_disPerforming Sobel operator processing in horizontal and vertical directions to obtain R_disRespectively, as Z, in the horizontal direction and in the vertical direction_h,r2And Z_v,r2Then calculating R_disIs marked as Z_r2，Wherein Z is_r2(x, y) represents Z_r2Gradient amplitude, Z, of pixel point with (x, y) as middle coordinate position_h,r2(x, y) represents Z_h,r2The horizontal gradient value Z of the pixel point with the (x, y) middle coordinate position_v,r2(x, y) represents Z_v,r2The gradient value in the vertical direction of the pixel point with the middle coordinate position (x, y) is that x is more than or equal to 1 and less than or equal to W ', y is more than or equal to 1 and less than or equal to H ', wherein W ' represents Z_r2Width of (A), H' represents Z_r2High of (d);

b3, calculating the threshold T' needed when dividing the area, <math> <mrow> <msup> <mi>T</mi> <mo>&prime;</mo> </msup> <mo>=</mo> <mi>&alpha;</mi> <mo>&times;</mo> <mfrac> <mn>1</mn> <mrow> <msup> <mi>W</mi> <mo>&prime;</mo> </msup> <mo>&times;</mo> <msup> <mi>H</mi> <mo>&prime;</mo> </msup> </mrow> </mfrac> <mo>&times;</mo> <mrow> <mo>(</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>W</mi> <mo>&prime;</mo> </msup> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>H</mi> <mo>&prime;</mo> </msup> </munderover> <msub> <mi>Z</mi> <mrow> <mi>r</mi> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>+</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>x</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>W</mi> <mo>&prime;</mo> </msup> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>=</mo> <mn>0</mn> </mrow> <msup> <mi>H</mi> <mo>&prime;</mo> </msup> </munderover> <msub> <mi>Z</mi> <mrow> <mi>r</mi> <mn>2</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math> wherein α is a constant, Z_r1(x, y) represents Z_r1Gradient amplitude, Z, of pixel point with (x, y) as middle coordinate position_r2(x, y) represents Z_r2The gradient amplitude of the pixel point with the middle coordinate position (x, y);

② b4, mixing Z_r1The gradient amplitude of the pixel point with the middle coordinate position (i, j) is recorded as Z_r1(i, j) reacting Z_r2The gradient amplitude of the pixel point with the middle coordinate position (i, j) is recorded as Z_r2(i, j), judgment of Z_r1(i,j)>T or Z_r2(i,j)>Whether T is true, and if so, determining R_orgAnd R_disThe pixel point with the middle coordinate position (i, j) belongs to the sensitive area, and order A_R(i, j) ═ 1, otherwise, R is determined_orgAnd R_disThe pixel point with the middle coordinate position (i, j) belongs to the non-sensitive area, and order A_R(i, j) ≦ 0, where i is 0 ≦ i ≦ W-8, and j is 0 ≦ H-8.

3. The method according to claim 1 or 2, wherein the step (c) is performed according to a preset algorithmMiddle beta₁The acquisition process comprises the following steps:

seventhly-1, adopting n undistorted stereo images to establish a distorted stereo image set under different distortion types and different distortion degrees, wherein the distorted stereo image set comprises a plurality of distorted stereo images, and n is more than or equal to 1;

obtaining an average subjective score difference value of each distorted stereo image in the distorted stereo image set by using a subjective quality evaluation method, and recording the average subjective score difference value as DMOS (diffused metal oxide semiconductor), wherein DMOS is 100-MOS (metal oxide semiconductor), MOS represents a subjective score mean value, and DMOS belongs to [0,100 ];

and (c) calculating an evaluation value Q of a sensitive region of the left viewpoint image of each distorted stereoscopic image in the distorted stereoscopic image set relative to the left viewpoint image of the corresponding undistorted stereoscopic image according to the operation processes from the step (i) to the step (c)_m,LAnd evaluation value Q of non-sensitive region_nm,L；

Seventhly-4, fitting the mean subjective score difference DMOS of the distorted stereo image in the set by adopting a mathematical fitting method and the corresponding Q_m,LAnd Q_nm,LThereby obtaining beta₁The value is obtained.