CN102663764A - Image quality evaluation method based on structural distortion and spatial frequency index - Google Patents

Abstract

The invention discloses an image quality evaluation method based on structural distortion and a spatial frequency index. An image quality evaluation method combining the structural distortion and the spatial frequency index is proposed on the basis of the image quality evaluation method based on the structural distortion, so that the image quality can be evaluated from the main factors influencing the image quality evaluation such as brightness, clarity and relevance. According to the image quality evaluation method based on the structural distortion and the spatial frequency index, consistency with the subjective feeling of the eye of the human beings can be realized when the image is objectively evaluated; and moreover since the selected calculation formula is simple, only the calculation between an original image and the distorted image pixels is involved, so that the method can quickly and efficiently evaluate the image quality.

Description

Image quality evaluating method based on structure distortion and spatial frequency index

Technical field

The invention belongs to the image quality evaluation field, particularly relate to the objective image quality evaluating method of a kind of brightness, sharpness and the degree of correlation based on image.

Background technology

Image quality evaluation is one of gordian technique of image processing system, is a kind of psychological activity of complicacy to the quality evaluation of image, has several factors to have influence on the judgement to the picture quality quality.If these factors are extracted respectively and are applied to well judging picture quality in the objective evaluation of image.These factors mainly contain: 1. brightness, appropriate brightness are the pacing itemss that the people observes image, and brightness is crossed by force or a little less than crossing all can cause decrease in image quality; 2. sharpness, people are commonly used fuzzy, the clear quality that waits vocabulary to describe image.See from the frequency domain angle, have sensation clearly when having only low-and high-frequency information ratio when piece image suitable; 3. the degree of correlation, i.e. similarity degree between the image.

Image quality evaluating method can be divided into two types of subjective quality evaluation method and method for evaluating objective quality.The subjective quality evaluation method is meant and lets the observer according to some opinion scales that provide in advance or own existing experience, treats test pattern and propose the quality judgement by vision or subjective impression, and provide quality score.The most frequently used subjective quality evaluation method has subjective average mark (MOS), and this method is used for many years.Yet the MOS method often receives the influence of the factor of observer own, and carries out visual psychology test and often need the time of cost longer, and environment of observation is had certain restriction.Therefore, in image quality evaluation, mainly use method for evaluating objective quality at present.This almost all has argumentation at all in about the document of image quality evaluation and research paper.The method for evaluating objective quality of image is meant that the error that departs from original image with reproduced image weighs the quality of reproduced image.It mainly is that applied mathematical model is represented the subjective feeling of vision to image.The application of mathematical model makes the characteristics that the evaluating objective quality of image has fast, stablizes, is easy to be quantized; The traditional image evaluating objective quality usually based on the serious more thought of the big more quality degradation of gray difference of standard picture, representative method has square error (MSE), Y-PSNR (PSNR), image definition, information entropy (H) etc.Though these method for evaluating objective quality commonly used seem simple, intuitive, mathematical expression is strict, and its evaluation result is often inconsistent with people's subjective sensation.Main cause is that PSNR value and MSE value all are the distortions from information data, does not consider that human eye has different visual experience to same information data distortion level, does not consider the spatial relation of information data yet.

Scholars such as Zhou Wang have proposed a kind of image quality evaluating method based on structure distortion.This image quality evaluation index is modeled as structure distortion rather than error with the picture quality reduction, comes assess image quality through the similarity of calculating original image and distorted image structure.This method is a simple and effective quality assessment algorithm; The advantage of this evaluation model is that factors such as the correlativity in the image processing process, luminance distortion and contrast are analyzed and researched, and these factors just people's evaluation map as the important evidence of output quality.The dynamic range of the image quality evaluation index Q of structure distortion is [1,1].Have only the i=1 of working as, 2 ..., N has y _i=x _iThe time, Q just can reach optimum value 1.Yet, when estimating the distorted image quality, certain defective is arranged also based on the image quality evaluation model of structure distortion.Its main cause is that this model do not consider that CSF among the human visual system is to the influence of picture appraisal.Human visual system's CSF mainly is reflected in the aspects such as sharpness of image.

The sharpness of image is meant the readability of picture engraving characteristic, just the contrast of characteristic (edge) and its background area.Sharpness is the tolerance that is used for reflecting the image border or puts the diffusion of characteristic, and diffusion is big more, and sharpness is low more, and diffusion is more little, and sharpness is high more.The index of quantitative evaluation sharpness is more.The spatial frequency index is through computer memory line frequency and space row frequency, and comprehensively both obtain the overall space frequency.The overall active degree of spatial frequency index reflected appraisal image space can reflect the sharpness of image effectively.But, for the distorted image of introducing noise, be not the image definition value subjective vision of hi-vision is good more more, reason is that the sharpness of image is that it does not consider the correlativity of distorted image and original image to the calculating between the pixel of image own.

Summary of the invention

In order to overcome the deficiency of above-mentioned prior art, the present invention provides a kind of image quality evaluating method based on structure distortion and spatial frequency index.

Technical scheme of the present invention is a kind of image quality evaluating method based on structure distortion and spatial frequency index, it is characterized in that being, may further comprise the steps:

Step 1; Calculating the average

of original image and distorted image utilizes relational expression

to calculate the mean flow rate degree of approximation of original image and distorted image

Wherein, x, y represent original image and distorted image respectively, and horizontal and vertical number of pixels is respectively m and n in the image,

$\stackrel{\‾}{x}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}x(i,j)$

$\stackrel{\‾}{y}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}y(i,j)$

X (i, j) and y (i j) is respectively original image and the distorted image gray-scale value at the capable j of i row;

Step 2, the square error σ of calculating original image and distorted image _x, σ _yAnd covariance sigma _Xy, utilize relational expression

Calculate the linear dependence degree of original image and distorted image,

Wherein, x, y represent original image and distorted image respectively, and image is respectively m and n at horizontal and vertical number of pixels,

${\mathrm{\σ}}_x={\left[\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(x(i,j)-\stackrel{\‾}{x})}^2\right]}^{1/2}$

${\mathrm{\σ}}_y={\left[\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-\stackrel{\‾}{y})}^2\right]}^{1/2}$

${\mathrm{\σ}}_{\mathrm{xy}}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}(x(i,j)-\stackrel{\‾}{x})(y(i,j)-\stackrel{\‾}{y})$

X (i, j) and y (i j) is respectively original image and the distorted image gray-scale value at the capable j of i row;

Step 3 is according to the square error σ of step 2 gained original image and distorted image _x, σ _y, utilize relational expression

Calculate the similarity of original image and distorted image;

Step 4; The spatial row frequency RF of calculated distortion image and space row frequency CF, following through the formula of spatial row frequency RF and space row frequency CF calculated distortion image space Frequency Index computer memory line frequency RF and space row frequency CF

$\mathrm{RF}=\sqrt{\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=2}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i,j-1))}^2}$

$\mathrm{CF}=\sqrt{\frac{1}{m\×n}\underset{i=2}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i-1,j))}^2}$

Wherein, m, n are respectively image at horizontal and vertical number of pixels, and (i j) is the gray-scale value of distorted image at the capable j row of i to y;

Step 5; Comprehensive step 1 averaging of income brightness degree of approximation, step 2 gained linear dependence degree, step 3 gained similarity and step 4 gained distorted image spatial frequency index are calculated based on the image quality evaluation index

of structure distortion and spatial frequency index and are carried out comprehensive evaluation according to calculating gained image quality evaluation index value from the tripartite picture quality of facing of brightness, sharpness and the correlativity of image.

Image quality evaluating method based on structure distortion provided by the present invention is modeled as structure distortion rather than error with the picture quality reduction.The present invention is on the basis based on the image quality evaluating method of structure distortion; Propose to combine the image quality evaluating method of spatial frequency index and structure distortion, thereby realized from the principal element that influences image quality evaluation being that brightness, sharpness and the degree of correlation are estimated picture quality.Experimental result shows, more can be consistent with the human eye subjective sensation when the objective evaluation image based on the image quality evaluating method of structure distortion and spatial frequency index.And, only relate to the calculating between original image and the distorted image pixel, so this algorithm can be estimated picture quality fast and effectively because the calculation relational expression of being selected for use is comparatively simple.

Description of drawings

Fig. 1 is the process flow diagram of the embodiment of the invention.

Fig. 2 is the used Lena test pattern of the test of the embodiment of the invention; Wherein Fig. 2-1 is an original image, and Fig. 2-2 is for adding the image of even distribution random noise, and Fig. 2-3 is for adding the image of Gaussian noise; Fig. 2-4 is the image of Fuzzy Processing, and Fig. 2-5 is the image of processed compressed.

Embodiment

The traditional image quality evaluating method can be divided into subjective picture quality evaluation method and objective image quality evaluating method two big classes.Traditional objective image quality evaluating method; Differentiate the quality of picture quality through the difference of each pixel and former pixel like square error (MSE), Y-PSNR methods such as (PSNR); Though mathematical expression is simple, convenience of calculation, its result is always not consistent with people's subjective feeling.And subjective evaluation method, can't be described by applied mathematical model through observer's scoring normalization being judged picture quality like subjective quality point system (MOS), in practical applications, wastes time and energy, and is difficult in the practical application and adopts.

The people is a kind of psychological activity of complicacy to image quality evaluation, many related factors is arranged to the judgement of people to the picture quality quality.If these factors are extracted the judgement that can perform well at last in addition comprehensively picture quality respectively.These factors mainly contain: 1. brightness, appropriate brightness are the pacing itemss that the people observes image, and brightness is crossed by force or a little less than crossing all can cause decrease in image quality; 2. sharpness, people are commonly used fuzzy, and the impression of subjective vision to picture quality described in the clear vocabulary that waits.See from the frequency domain angle, the reason that the high fdrequency component deficiency is fuzzy often, and high fdrequency component too much can cause the coarse of image.Have sensation clearly when having only low-and high-frequency information ratio when piece image suitable; 3. the degree of correlation, i.e. similarity degree between the image.Similarity degree can not be interpreted as the difference of pixel grey scale simply, but the form of image, similar content.

Image quality evaluating method based on structure distortion.This image quality evaluation index is modeled as structure distortion rather than error with the picture quality reduction, comes assess image quality through the similarity of calculating original image and distorted image structure.This method is a simple and effective quality assessment algorithm; The advantage of this evaluation model is that factors such as the correlativity in the image processing process, luminance distortion and contrast are analyzed and researched, and these factors just people's evaluation map as the important evidence of output quality.Yet, when estimating the distorted image quality, certain defective is arranged also based on the image quality evaluation model of structure distortion.Its main cause is that this model do not consider that CSF among the human visual system is to the influence of picture appraisal.Human visual system's CSF mainly is reflected in the aspects such as sharpness of image.

Based on the image quality evaluating method of structure distortion and spatial frequency index on basis based on the image quality evaluating method of structure distortion; Introduce the spatial frequency index image definition is estimated, thereby realized from brightness, sharpness, the tripartite quality assessment of the degree of correlation in the face of image.The result that this method draws more meets the visual experience of human eye.

Technical scheme of the present invention can adopt computer software technology to realize automatic operational scheme, specifies technical scheme of the present invention below in conjunction with embodiment.Like Fig. 1, the flow process of the embodiment of the invention may further comprise the steps:

Step 1; Calculating the average

of original image and distorted image utilizes relational expression

to calculate the mean flow rate degree of approximation of original image and distorted image

Wherein, x, y represent original image and distorted image respectively, and horizontal and vertical number of pixels is respectively m and n in the image,

$\stackrel{\‾}{x}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}x(i,j)$

$\stackrel{\‾}{y}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}y(i,j)$

X (i; J) and y (i; J) be respectively original image and the distorted image gray-scale value at the capable j row of i, the span of mean flow rate degree of approximation is [0,1]; When having only as

, mean flow rate degree of approximation value is 1;

Step 2, the square error σ of calculating original image and distorted image _x, σ _yAnd covariance sigma _Xy, utilize relational expression

Calculate the linear dependence degree of original image and distorted image,

Wherein, x, y represent original image and distorted image respectively, and image is respectively m and n at horizontal and vertical number of pixels,

${\mathrm{\σ}}_x={\left[\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(x(i,j)-\stackrel{\‾}{x})}^2\right]}^{1/2}$

${\mathrm{\σ}}_y={\left[\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-\stackrel{\‾}{y})}^2\right]}^{1/2}$

${\mathrm{\σ}}_{\mathrm{xy}}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}(x(i,j)-\stackrel{\‾}{x})(y(i,j)-\stackrel{\‾}{y})$

X (i, j) and y (i j) is respectively original image and the distorted image gray-scale value at the capable j of i row, and the dynamic range of linear dependence degree is [1,1];

Step 3 is according to the square error σ of step 2 gained original image and distorted image _x, σ _y, utilize relational expression

Calculate the similarity of original image and distorted image;

The span of similarity is [0,1], has only the σ of working as _x=σ _yThe time, obtain optimum value 1;

Step 4; The spatial row frequency RF of calculated distortion image and space row frequency CF, following through the formula of spatial row frequency RF and space row frequency CF calculated distortion image space Frequency Index

computer memory line frequency RF and space row frequency CF

$\mathrm{RF}=\sqrt{\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=2}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i,j-1))}^2}$

$\mathrm{CF}=\sqrt{\frac{1}{m\×n}\underset{i=2}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i-1,j))}^2}$

Wherein, m, n are respectively image at horizontal and vertical number of pixels, y (i; J) be the gray-scale value of distorted image at the capable j row of i, (i-1 j) is the gray-scale value of distorted image at the capable j row of i-1 to y; (i is the gray-scale value of distorted image at the capable j-1 row of i j-1) to y, and the codomain of image space Frequency Index is [0; 255], its value is big more, and the sharpness of image is high more;

Step 5; Comprehensive step 1 averaging of income brightness degree of approximation, step 2 gained linear dependence degree, step 3 gained similarity and step 4 gained distorted image spatial frequency index; Calculating is [0 based on the codomain of image quality evaluation index

the image quality evaluation index of structure distortion and spatial frequency index; 255], carry out comprehensive evaluation from the tripartite picture quality of facing of brightness, sharpness and the correlativity of image according to calculating gained image quality evaluation index value.

During practical implementation, preceding 4 steps can parallel processing, and those skilled in the art can adjust voluntarily.But for the purpose of economizing on resources, step 3 can directly be utilized the square error σ of step 2 gained original image and distorted image _x, σ _y, therefore be adapted at carrying out after the step 2.

For the purpose of explanation effect of the present invention, adopt software to make an experiment according to embodiment of the invention design:

(1) utilize software to read in original image and distorted image; Experiment selected Lena figure is a test pattern; Original image is designated as Fig. 2-1, and the image that adds even distribution random noise is designated as Fig. 2-2, and the image that adds Gaussian noise is designated as Fig. 2-3; The image of Fuzzy Processing is designated as Fig. 2-4, and the image of processed compressed is designated as Fig. 2-5.At this moment, image with the stored in form of matrix, is m * n if establish the image size in software, and then the size of matrix is m * n, in this test, makes N=m * n, and each element in the matrix is being stored the pixel value of correspondence image.

(2) in the following steps, make x, y represents original image and distorted image respectively, for each width of cloth distorted image, compares with original image respectively, and it is corresponding to calculate original image and distorted image

σ _x, σ _y, σ _Xy, through formula

Obtain the Q value that each width of cloth distorted image is compared with original image.The Q value has reflected the index of mean flow rate degree of approximation, linear dependence degree and the similarity of image.Wherein, make Q21, Q31, Q41, Q51 represent distorted image 2-2 respectively, 2-3,2-4,2-5 compare the Q value that calculates with original image 2-1, in this test, Q21=0.647, Q 31=0.647, Q 41=0.647, Q 51=0.647.

(3) programming calculated distortion image space line frequency And space row frequency $\mathrm{CF}=\sqrt{\frac{1}{m\×n}\underset{i=2}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i-1,j))}^2},$ Pass through relational expression $\mathrm{SF}=\sqrt{{\mathrm{RF}}^2+{\mathrm{CF}}^2}$ The computer memory Frequency Index, in the formula, m, n are respectively the horizontal and vertical number of pixels of distorted image, (i j) is the gray-scale value of distorted image at the capable j row of i to y.Wherein, make SF21, SF31, SF41, SF51 represent distorted image 2-2 respectively, 2-3,2-4, the P value that 2-5 computer memory Frequency Index obtains, in this test, SF21=31.535, SF31=29.417, SF41=9.998, SF51=14.367.

(4) the present invention sets the image quality evaluation index based on structure distortion and spatial frequency index:

SFQ can carry out comprehensive evaluation from the tripartite picture quality of facing of brightness, sharpness and the correlativity of image.Make SFQ21, SFQ31, SFQ41, SFQ51 represent distorted image 2-2 respectively, 2-3,2-4,2-5 compare the PQ value that calculates with original image 2-1, in this test, SFQ21=20.403, SFQ31=19.033, SFQ41=6.479, SFQ51=9.295.

(5) this test gives distorted image 2-2; 2-3,2-4,2-5 compare the traditional images quality assessment parameter PSNR value that calculates with original image 2-1; The PSNR value draws

wherein by following formula; M * n is the size of image, and x (i, j); (i j) represents the respective pixel value of original image and distorted image respectively to y.PSNR21＝27.587，PSNR31＝22.885，PSNR41＝24.84，PSNR51＝25.523。

(6) can obviously find out, Fig. 2-2,2-3,2-4, the visual effect of 2-5 is also inconsistent, and the subjective sensation of Fig. 2-2 and Fig. 2-3 will obviously be superior to Fig. 2-4 and Fig. 2-5.SFQ value in the subordinate list 1 has just in time reflected this point, however Q21=Q31=Q41=Q51=0.647.Q value and PSNR value can not embody the difference of each distorted image, so the result who draws based on the image quality evaluating method of structure distortion and spatial frequency index more meets the visual experience of human eye.

Subordinate list 1Lena figure Q value, PSNR value, SFQ value test result

The present invention also can adopt the software modularity designing technique to be embodied as the image quality evaluation system based on structure distortion and spatial frequency index, comprising:

Mean flow rate degree of approximation module; The average

that is used to calculate original image and distorted image utilizes relational expression

to calculate the mean flow rate degree of approximation of original image and distorted image, calculates averaging of income brightness degree of approximation input comprehensive evaluation module;

Linear dependence degree module is used to calculate the square error σ of original image and distorted image _x, σ _yAnd factor sigma _Xy, utilize relational expression Calculate the linear dependence degree of original image and distorted image, calculate gained linear dependence degree input comprehensive evaluation module; The similarity module is used to calculate the square error σ of original image and distorted image _x, σ _y, utilize relational expression

Calculate the similarity of original image and distorted image; Calculate gained similarity input comprehensive evaluation module;

Distorted image spatial frequency index module; The spatial row frequency RF and the space row frequency CF that are used for the calculated distortion image calculate gained distorted image spatial frequency index input comprehensive evaluation module through spatial row frequency RF and space row frequency CF calculated distortion image space Frequency Index

;

The comprehensive evaluation module; Be used for comprehensive mean flow rate degree of approximation, linear dependence degree, similarity and distorted image spatial frequency index, calculate based on the image quality evaluation index

of structure distortion and spatial frequency index and carry out comprehensive evaluation from the tripartite picture quality of facing of brightness, sharpness and the correlativity of image according to calculating gained image quality evaluation index value.

Specific embodiment described herein only is that the present invention's spirit is illustrated.Person of ordinary skill in the field of the present invention can make various modifications or replenishes or adopt similar mode to substitute described specific embodiment, but can't depart from spirit of the present invention or surmount the defined scope of appended claims.

Claims (1)

1. the image quality evaluating method based on structure distortion and spatial frequency index is characterized in that being, may further comprise the steps:

Step 1; Calculating the average

of original image and distorted image utilizes relational expression

to calculate the mean flow rate degree of approximation of original image and distorted image

Wherein, x, y represent original image and distorted image respectively, and horizontal and vertical number of pixels is respectively m and n in the image,

$\stackrel{\‾}{x}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}x(i,j)$

$\stackrel{\‾}{y}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}y(i,j)$

X (i, j) and y (i j) is respectively original image and the distorted image gray-scale value at the capable j of i row;

Step 2, the square error σ of calculating original image and distorted image _x, σ _yAnd covariance sigma _Xy, utilize relational expression

Calculate the linear dependence degree of original image and distorted image,

Wherein, x, y represent original image and distorted image respectively, and image is respectively m and n at horizontal and vertical number of pixels,

${\mathrm{\σ}}_x={\left[\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(x(i,j)-\stackrel{\‾}{x})}^2\right]}^{1/2}$

${\mathrm{\σ}}_y={\left[\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-\stackrel{\‾}{y})}^2\right]}^{1/2}$

${\mathrm{\σ}}_{\mathrm{xy}}=\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}(x(i,j)-\stackrel{\‾}{x})(y(i,j)-\stackrel{\‾}{y})$

X (i, j) and y (i j) is respectively original image and the distorted image gray-scale value at the capable j of i row;

Step 3 is according to the square error σ of step 2 gained original image and distorted image _x, σ _y, utilize relational expression

Calculate the similarity of original image and distorted image;

Step 4; The spatial row frequency RF of calculated distortion image and space row frequency CF, following through the formula of spatial row frequency RF and space row frequency CF calculated distortion image space Frequency Index

computer memory line frequency RF and space row frequency CF

$\mathrm{RF}=\sqrt{\frac{1}{m\×n}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=2}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i,j-1))}^2}$

$\mathrm{CF}=\sqrt{\frac{1}{m\×n}\underset{i=2}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{(y(i,j)-y(i-1,j))}^2}$

Wherein, m, n are respectively image at horizontal and vertical number of pixels, and (i j) is the gray-scale value of distorted image at the capable j row of i to y;

Step 5; Comprehensive step 1 averaging of income brightness degree of approximation, step 2 gained linear dependence degree, step 3 gained similarity and step 4 gained distorted image spatial frequency index are calculated based on the image quality evaluation index of structure distortion and spatial frequency index and are carried out comprehensive evaluation according to calculating gained image quality evaluation index value from the tripartite picture quality of facing of brightness, sharpness and the correlativity of image.

Images