CN101833766A - Stereo image objective quality evaluation algorithm based on GSSIM - Google Patents

Abstract

The invention belongs to the image processing field and relates to a stereo image quality evaluation method based on GSSIM. The method includes the steps: (1) for left image and right image, gradient structure similarity values are respectively solved, the average of the two values is solved, so as to obtain stereo image quality evaluation value QE; (2) the following method is adopted to carrying out image stereoscopic perception objective evaluation: absolute difference image of the original image and processed image is calculated, Mu1 and Mu2 of the absolute difference image are solved; Sobel operator is used for solving the gradient amplitude image of the absolute difference image; filtering is carried out on the absolute difference image of the original image, and original image binocular parallax distribution situation is calculated; Dl(x, y), Dcg(x, y) and Dsg(x, y) values at binocular parallax position are solved; DSSIM value at the binocular parallax position is calculated; and image gradient structure similarity namely image stereoscopic perception objective evaluation value DE is calculated. The invention can be well applied to stereo image quality evaluation, and correlation of objective evaluation result and subjective evaluation result is strong.

Description

Three-dimensional image objective quality evaluation algorithm based on GSSIM

Technical field

The invention belongs to image processing field, relate to a kind of three-dimensional image objective quality evaluation algorithm.

Background technology

The quality assessment of stereoscopic image is a very difficult job, and existing plane picture evaluating objective quality algorithm concerns with decoded picture quality before can only estimating same viewpoint coding usually.There is the high correlation of human eye standard parallax in stereo-picture between adjacent viewpoint, if two adjacent viewpoint picture quality separately is all than higher, parallax composition minimizing even anti-phase between viewpoint, the stereoscopic sensation that human eye can be enjoyed will be had a greatly reduced quality, so existing plane picture evaluating objective quality algorithm can not replace the three-dimensional image objective quality evaluation algorithm.If a kind of compressed encoding causes relief heavy losses even causes beholder's kopiopia, be not suitable for the application of stereo-picture.Therefore, the necessary objective evaluation standard of setting up a stereo-picture.

Gradient-structure similarity (GSSIM) evaluation method is a kind of plane picture method for evaluating objective quality, discover that human eye is the most responsive and payes attention to for edge of image and texture structure information, the structural information of edge and texture probably is the most important parts of picture structure information in other words, and gradient can be reacted the edge of image texture information preferably, so adopt the gradient-structure similarity also can carry out the stereo image quality evaluation.This method is as follows:

The extraction of gradient information: determine that by the calculating of gradient image edge information is the most common and effective and efficient manner.Adopt the Sobel operator that image is carried out the calculating of gradient, as shown in Figure 1.

Each pixel x for image X _{I, j}(i, j represent horizontal ordinate value) can pass through Sobel operator definitions its " gradient information vector "; V _{I, j}={ dx _{I, j}, dy _{I, j}Wherein, dx _{I, j}And dy _{I, j}Horizontal edge operator H and vertical edge operator v by Fig. 1 draws respectively, easy for algorithm, and the gradient magnitude that defines image pixel approx is:

Amp _i，j＝|dx _i，j|+|dy _i，j|

Accordingly, the gradient direction of each pixel is defined as:

Utilize the structural similarity (GSSIM) of the amplitude information proposition of image:, can obtain the gradient magnitude of each pixel of image, thereby obtain and image X and corresponding gradient image X ' of Y and Y ' with the gradient magnitude of Sobel operator and image pixel based on gradient.Therefore, sub-piece gradient contrast relatively may be defined as:

$C_g(x,y)=\frac{2{\mathrm{\&sigma;}}_{x^{\&prime;}}{\mathrm{\&sigma;}}_{y^{\&prime;}}+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}{{\mathrm{\&sigma;}}_{x^{\&prime;}}^2+{\mathrm{\&sigma;}}_{y^{\&prime;}}^2+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}$

σ is a standard deviation, C ₂For adjusting parameter.

The texture ratio of sub-piece gradient may be defined as:

$S_g(x,y)=\frac{{\mathrm{\&sigma;}}_{x^{\&prime;}{y}^{\&prime;}}+C_3}{{\mathrm{\&sigma;}}_{x^{\&prime;}}{\mathrm{\&sigma;}}_{y^{\&prime;}}+C_3}$

C ₃For adjusting parameter.

Structural similarity (GSSIM) may be defined as:

GSSIM(x，y)＝[1(x，y)] ^α[c _g(x，y)] ^β[s _g(x，y)] ^γ

α, beta, gamma are the weight exponential quantity.

For the comparison of the similarity of entire image, can draw by the similarity scoring of average each height piece:

Summary of the invention

The objective of the invention is to overcome the above-mentioned deficiency of prior art, propose a kind of three-dimensional image objective quality evaluation method.The present invention introduces gradient-structure similarity (GSSIM) evaluation method in the three-dimensional image objective quality evaluation, carries out image quality evaluation (QE) and stereoscopic sensation evaluation (DE) by stereoscopic image, finishes the judge of stereoscopic image quality.The present invention adopts following technical scheme:

A kind of stereo image quality evaluation method based on GSSIM comprises the following steps:

(1) for left image and right image, utilize gradient-structure similarity GSSIM algorithm respectively, ask for gradient-structure similarity value, ask for the average of two values again, obtain stereo image quality evaluation of estimate QE;

(2) adopt following method to carry out image stereo perception objective evaluation:

The first step: the viewpoint of original image to (L1, R1) and the viewpoint of handling the back image to (L2 R2) carries out the phase reducing respectively, draws absolute difference image X1 and X2;

In second step, carry out luminance test, by the following μ that obtains absolute difference image X1 and X2 of formula ₁, μ ₂:

${\mathrm{\&mu;}}_{1_{\mathrm{mn}}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{x}_{1\mathrm{ij}},$

${\mathrm{\&mu;}}_{2_{\mathrm{mn}}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{x}_{2\mathrm{ij}}$

In the formula, w is a center symmetrical Gaussian weighting windows

The N value is 11;

In the 3rd step, obtain gradient magnitude image M 1 and the M2 of absolute difference image X1 and X2 with the Sobel operator, and obtain image M 1 and M2 brightness average μ _{1 '}, μ _{2 '}, again by the following standard deviation sigma of obtaining of formula _{1 '}, σ _{2 '}, and covariance sigma _{1 ' 2 '}:

${\mathrm{\&sigma;}}_{1_{\mathrm{mn}}^{\&prime;}}={\left(\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{(M_{1\mathrm{ij}}-{\mathrm{\&mu;}}_{1_{\mathrm{mn}}^{\&prime;}})}^2\right)}^{1/2}$

${\mathrm{\&sigma;}}_{2_{\mathrm{mn}}^{\&prime;}}={\left(\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{({\mathrm{<figure-callout\; id="2"\; label="M"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">M</figure-callout>}}_{2\mathrm{ij}}-{\mathrm{\&mu;}}_{2_{\mathrm{mn}}^{\&prime;}})}^2\right)}^{1/2}$

${\mathrm{\&sigma;}}_{1^{\&prime;}{2}_{\mathrm{mn}}^{\&prime;}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}({\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">M</figure-callout>}}_{1\mathrm{ij}}-{\mathrm{\&mu;}}_{1_{\mathrm{mn}}^{\&prime;}})({\mathrm{<figure-callout\; id="2"\; label="M"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">M</figure-callout>}}_{2\mathrm{ij}}-{\mathrm{\&mu;}}_{2_{\mathrm{mn}}^{\&prime;}})$

The 4th step, the absolute difference image X1 of original image is carried out filtering, threshold value decision operation, calculate former figure binocular parallax distribution situation;

In the 5th step, use μ ₁, μ ₂, σ _{1 '}, σ _{2 '}With σ _{1 ' 2 '}, in conjunction with the former figure parallax distribution situation that calculates, obtain the binocular parallax place Dl (x, y), Dcg (x, y) with Dsg (x, y) value, computing formula is as follows:

$D1(x,y)=\frac{2{\mathrm{\&mu;}}_1{\mathrm{\&mu;}}_2+{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_1}{{\mathrm{\&mu;}}_1^2+{\mathrm{\&mu;}}_2^2+{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_1}$

${\mathrm{Dc}}_g(x,y)=\frac{2{\mathrm{\&sigma;}}_{1^{\&prime;}}{\mathrm{\&sigma;}}_{2^{\&prime;}}+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}{{\mathrm{\&sigma;}}_{1^{\&prime;}}^2+{\mathrm{\&sigma;}}_{2^{\&prime;}}^2+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}$

${\mathrm{Ds}}_g(x,y)=\frac{{\mathrm{\&sigma;}}_{1^{\&prime;}{2}^{\&prime;}}+C_3}{{\mathrm{\&sigma;}}_{1^{\&prime;}{2}^{\&prime;}}+C_3};$

The 6th step, by formula DSSIM (x, y)=[Dl (x, y)] ^α[Dc _g(x, y)] ^β[Ds _g(x, y)] ^γ(3-34) calculate the DSSIM value at binocular parallax place, α wherein, the beta, gamma value all is 1;

In the 7th step, by the DSSIM value, utilize the image gradient structural similarity DMSSIM of column count binocular parallax place under the formula, i.e. image stereo perception objective evaluation value DE:

In the 8th step, the stereoscopic image quality is estimated: the QE value is big more, and picture quality is good more, and the DE value is big more, and stereoscopic sensation is good more.

The three-dimensional image objective quality evaluation method that uses the present invention to propose, can carry out quality evaluation from picture quality and two aspect stereoscopic image of stereoscopic sensation, experimental results demonstrate that this algorithm can perform well in the quality assessment of stereo-picture, objective evaluation result and subjective assessment results relevance are very strong, and effectively stereoscopic image is made evaluation.

Description of drawings

Fig. 1 Sobel operator (a) is vertical edge operator V, (b) is horizontal edge operator H.

Fig. 2 stereo perception objective evaluation of the present invention flow process

Fig. 2 stereo perception objective evaluation flow process.

Embodiment

The index of stereoscopic image objective evaluation proposed by the invention is divided into two parts: one is the image evaluating objective quality, and one is the stereoscopic sensation objective evaluation.

1, image evaluating objective quality (quality evaluation:QE)

Continue to use gradient-structure similarity (GSSIM) evaluation method for the image evaluating objective quality.

Three-dimensional image objective quality evaluation value of the present invention be left image GMSSIM value and right image GMSSIM value average promptly: QE=1-(GMSSIM _{A left side}+ GMSSIM _Right)/2

2, image stereo perception objective evaluation (3D sense evaluation:DE)

Image stereo perception objective evaluation algorithm of the present invention is continued to use the evaluating of GSSIM algorithm, and viewpoint organically is combined in the evaluating absolute difference figure.Stereo perception objective evaluation flow process such as Fig. 2:

At first, original viewpoint to (L1, R1) and handle backsight point to (L2 R2) carries out the phase reducing respectively, draws absolute difference image X1 and X2;

X1＝|L1-R1|

X2＝|L2-R2|

In second step, carry out luminance test, by the following μ that obtains absolute difference image X1 and X2 of formula ₁, μ ₂

${\mathrm{\&mu;}}_{1_{\mathrm{mn}}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{x}_{1\mathrm{ij}}$

${\mathrm{\&mu;}}_{2_{\mathrm{mn}}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{x}_{2\mathrm{ij}}$

W is a center symmetrical Gaussian weighting windows

The N value is 11.

In the 3rd step, obtain gradient magnitude image M 1 and the M2 of absolute difference image X1 and X2 with the Sobel operator, and obtain image M 1 and M2 brightness average μ _{1 '}, μ _{2 '}, again by the following standard deviation sigma of obtaining of formula _{1 '}, σ _{2 '}, and covariance sigma 1 ' 2 '.

${\mathrm{\&sigma;}}_{1_{\mathrm{mn}}^{\&prime;}}={\left(\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{(M_{1\mathrm{ij}}-{\mathrm{\&mu;}}_{1_{\mathrm{mn}}^{\&prime;}})}^2\right)}^{1/2}$

${\mathrm{\&sigma;}}_{2_{\mathrm{mn}}^{\&prime;}}={\left(\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{({\mathrm{<figure-callout\; id="2"\; label="M"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">M</figure-callout>}}_{2\mathrm{ij}}-{\mathrm{\&mu;}}_{2_{\mathrm{mn}}^{\&prime;}})}^2\right)}^{1/2}$

${\mathrm{\&sigma;}}_{1^{\&prime;}{2}_{\mathrm{mn}}^{\&prime;}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}({\mathrm{<figure-callout\; id="1"\; label="M"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">M</figure-callout>}}_{1\mathrm{ij}}-{\mathrm{\&mu;}}_{1_{\mathrm{mn}}^{\&prime;}})({\mathrm{<figure-callout\; id="2"\; label="M"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">M</figure-callout>}}_{2\mathrm{ij}}-{\mathrm{\&mu;}}_{2_{\mathrm{mn}}^{\&prime;}})$

The 4th step, the absolute difference image X1 of original image is carried out filtering, threshold value decision operation, remove faint noise and low amplitude value composition, reach to reduce and disturb purpose, the former figure binocular parallax distribution situation that calculates then.Because the existence of psychological stereoscopic vision, the people can feel that the scope of parallax of small magnitude is very little, and most of having been covered is so in this step, remove the influence of small magnitude parallax by the threshold value judgement.

In the 5th step, use μ ₁, μ ₂, σ _{1 '}, σ _{2 '}With σ _{1 ' 2 '}, in conjunction with the former figure parallax distribution situation that calculates, (x, y), (x, y) (x, y) value, computing formula be (3-31), (3-32) and (3-33) to Dcg with Dsg to obtain the Dl of binocular parallax place (being made as D).

$D1(x,y)=\frac{2{\mathrm{\&mu;}}_1{\mathrm{\&mu;}}_2+{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_1}{{\mathrm{\&mu;}}_1^2+{\mathrm{\&mu;}}_2^2+{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_1}$

${\mathrm{Dc}}_g(x,y)=\frac{2{\mathrm{\&sigma;}}_{1^{\&prime;}}{\mathrm{\&sigma;}}_{2^{\&prime;}}+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}{{\mathrm{\&sigma;}}_{1^{\&prime;}}^2+{\mathrm{\&sigma;}}_{2^{\&prime;}}^2+{\mathrm{<figure-callout\; id="2"\; label="C"\; filenames="HDA0000021233210000011.png"\; state="\{\{state\}\}">C</figure-callout>}}_2}$

${\mathrm{Ds}}_g(x,y)=\frac{{\mathrm{\&sigma;}}_{1^{\&prime;}{2}^{\&prime;}}+C_3}{{\mathrm{\&sigma;}}_{1^{\&prime;}{2}^{\&prime;}}+C_3}$

In the 6th step, (x, y), (x, y) (x, y) each value calculate the DSSIM value at binocular parallax place to Dcg by formula (3-34) with Dsg according to the Dl at binocular parallax place.

DSSIM(x，y)＝[Dl(x，y)] ^α[Dc _g(x，y)] ^β[Ds _g(x，y)] ^γ

α wherein, the beta, gamma value all is 1.

In the 7th step,, utilize the image gradient structural similarity DMSSIM of column count binocular parallax place under the formula, i.e. image stereo perception objective evaluation value---DE by DSSIM.

$\mathrm{DE}=\mathrm{DMSSIM}=1-\frac{1}{D}\underset{i=1}{\overset{D}{\mathrm{\&Sigma;}}}\mathrm{DSSIM}(x_i,y_i)$

Sum up:

When the QE value was 0.9-1, picture quality was fine;

When the QE value was 0.8-0.9, picture quality was better;

When the QE value is 0.55-0.8, picture quality one;

The QE value is 0.55 when following, and picture quality is very poor.

When the DE value was 0.85-1, stereoscopic sensation was fine;

When the DE value was 0.7-0.85, stereoscopic sensation was better;

When the DE value is 0.45-0.7, stereoscopic sensation one;

The DE value is 0.45 when following, and stereoscopic sensation is very poor.

The present invention is best to double vision point stereo image quality evaluation effect; If multi-viewpoint stereo image can be divided into a plurality of double vision point stereo-pictures and estimate.

Claims (1)

1. the stereo image quality evaluation method based on GSSIM comprises the following steps:

(1) for left image and right image, utilize gradient-structure similarity GSSIM algorithm respectively, ask for gradient-structure similarity value, ask for the average of two values again, obtain stereo image quality evaluation of estimate QE;

(2) adopt following method to carry out image stereo perception objective evaluation:

The first step: the viewpoint of original image to (L1, R1) and the viewpoint of handling the back image to (L2 R2) carries out the phase reducing respectively, draws absolute difference image X1 and X2;

In second step, carry out luminance test, by the following μ that obtains absolute difference image X1 and X2 of formula ₁, μ ₂:

In the formula, w is a center symmetrical Gaussian weighting windows

The N value is 11;

In the 3rd step, obtain gradient magnitude image M 1 and the M2 of absolute difference image X1 and X2 with the Sobel operator, and obtain image M 1 and M2 brightness average μ _{1 '}, μ _{2 '}, again by the following standard deviation sigma of obtaining of formula _{1 '}, σ _{2 '}, and covariance sigma _{1 ' 2 '}:

${\mathrm{\&sigma;}}_{1_{\mathrm{mn}}^{\&prime;}}={\left(\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{(M_{1\mathrm{ij}}-{\mathrm{\&mu;}}_{1_{\mathrm{mn}}^{\&prime;}})}^2\right)}^{1/2}$

${\mathrm{\&sigma;}}_{2_{\mathrm{mn}}^{\&prime;}}={\left(\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{(M_{2\mathrm{ij}}-{\mathrm{\&mu;}}_{2_{\mathrm{mn}}^{\&prime;}})}^2\right)}^{1/2}$

${\mathrm{\&sigma;}}_{1^{\&prime;}{2}_{\mathrm{mn}}^{\&prime;}}=\underset{i=m-\frac{N-1}{2}}{\overset{m+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}\underset{j=n-\frac{N-1}{2}}{\overset{n+\frac{N-1}{2}}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}(M_{1\mathrm{ij}}-{\mathrm{\&mu;}}_{1_{\mathrm{mn}}^{\&prime;}})(M_{2\mathrm{ij}}-{\mathrm{\&mu;}}_{2_{\mathrm{mn}}^{\&prime;}});$

The 4th step, the absolute difference image X1 of original image is carried out filtering, threshold value decision operation, calculate former figure binocular parallax distribution situation;

In the 5th step, use μ ₁, μ ₂, σ _{1 '}, σ _{2 '}With σ _{1 ' 2 '}, in conjunction with the former figure parallax distribution situation that calculates, obtain the binocular parallax place D1 (x, y), Dcg (x, y) with Dsg (x, y) value, computing formula is as follows:

$D1(x,y)=\frac{2{\mathrm{\&mu;}}_1{\mathrm{\&mu;}}_2+C_1}{{\mathrm{\&mu;}}_1^2+{\mathrm{\&mu;}}_2^2+C_1}$

${\mathrm{Dc}}_g(x,y)=\frac{2{\mathrm{\&sigma;}}_{1^{\&prime;}}{\mathrm{\&sigma;}}_{2^{\&prime;}}+C_2}{{\mathrm{\&sigma;}}_{1^{\&prime;}}^2+{\mathrm{\&sigma;}}_{2^{\&prime;}}^2+C_2}$

${\mathrm{Ds}}_g(x,y)=\frac{{\mathrm{\&sigma;}}_{1^{\&prime;}{2}^{\&prime;}}+C_3}{{\mathrm{\&sigma;}}_{1^{\&prime;}}{\mathrm{\&sigma;}}_{2^{\&prime;}}+C_3};$

The 6th step, by formula DSSIM (x, y)=[D1 (x, y)] ^α[Dc _g(x, y)] ^β[Ds _g(x, y)] ^γ(3-34) calculate the DSSIM value at binocular parallax place, α wherein, the beta, gamma value all is 1;

In the 7th step, by the DSSIM value, utilize the image gradient structural similarity DMSSIM of column count binocular parallax place under the formula, i.e. image stereo perception objective evaluation value DE:

In the 8th step, the stereoscopic image quality is estimated: the QE value is big more, and picture quality is good more, and the DE value is big more, and stereoscopic sensation is good more.

Images