CN104408707A - Rapid digital imaging fuzzy identification and restored image quality assessment method - Google Patents

Abstract

The invention discloses a rapid digital imaging fuzzy identification and restored image quality assessment method, and relates to a fuzzy identification and restored image quality assessment method, for solving the problems of poor real-time performance, incapability of effective identification of a blurred image and a clear image and effective assessment of an image restoration result, and instable output result after fuzzy restoration processing by use of a conventional method. The method is realized through the following steps: 1, obtaining a gradient image G (i, j); 2, calculating an average gray scale gradient value; 3, obtaining statistical distribution information of the gradient image; 4, performing threshold determining on BIM; 5, obtaining a restored image; 6, generating reference images F1 and D1; 7, calculating brightness similarities, contrast similarities and structural similarities between the reference images and an image to be assessed; 8, calculating g(F(i, j), F1) and g(D(i, j), D1); 9, obtaining a non-reference-image assessment index; 10, determining the non-reference-image assessment index; and the like. The rapid digital imaging fuzzy identification and restored image quality assessment method is applied to the field of fuzzy identification and restored image quality assessment.

Description

A kind of fast digital image blur is differentiated and restored image method for evaluating quality

Technical field

The present invention relates to image quality measure field, particularly fuzzy discriminating and restored image quality evaluation field.

Background technology

When digital camera imaging, relative motion between camera lens and imaging scene or camera lens defocus the image or video blur that all can cause collecting, cause that the contrast of image reduces, edge and interior details information weakens, affect picture quality, make the directly perceived and digital image information disposal system of vision be difficult to the area-of-interest detected exactly in image, have a strong impact on the analysis to obtained image and video information and understanding.Generally picture quality be can improve to a certain extent by digital blur image restoration system, a part of edge and image interior details information recovered.But under many circumstances, be limited by the shortage of image prior information and the defect of restoration algorithm, this recuperation may produce more artificial traces as ringing effect etc., cause the picture quality after digital restoration process not even as untreated image.Therefore, one is had to the digital image restoration system of real practicality, carrying out before and after digital restoration to image blur image, two critical processing links must be implemented, namely whether there is image blur to the image obtained to differentiate, after restoration disposal is carried out to blurred picture, carry out image quality evaluation simultaneously, after guaranteeing image restoration, obtain image or the video of better quality.Can aforementioned two information processing links effectively work and really apply real system for digital imagery smear restoration method and have vital role, in civilian intelligent video monitoring system, Intelligent traffic management systems etc. and Military Application as imaging reconnaissance system, space optical remote and missile imaging guidance field in all there is important using value.

In existing digital picture smear restoration system and the image blurring restored method studied, nearly all there is no this step of fuzzy discriminating, be all directly image is restored usually, following two problems can be there is like this:

1) in real time video processing, the image of each frame only has the extremely short processing time by frame per second restriction, and the general frame per second of civil area is between 20 ~ 30, and namely every frame only has the processing time of 30 ~ 50ms, and military domain frame per second is 50 ~ 60, every frame only has the processing time of 15 ~ 20ms.And existing restoration algorithm adopts the mode of iteration to carry out mostly, poor real, even if optimize processing time of the good every frame of the video of restoration algorithm to same size also close to 40ms, in practical application, image blurring recovery is generally just as pre-service link, also to carry out as target detection in the processing time of each two field picture, identify, the tasks such as tracking, if carry out fuzzy discriminating, image blurring recovery time can be used for follow-up work, be more conducive to the realization (instead of improving single frames processing speed) of real time video processing task, if do not add differentiation just to carry out processing the significant wastage that will cause computational resource to image, be difficult to reach requirement of real-time.

2) in the Video processing of reality, not each two field picture is all fuzzy, for picture rich in detail, if carry out smear restoration process, can introduce as part artificial traces such as ringing effects on the contrary, picture frame quality after process declines on the contrary relative to time untreated, has a strong impact on the validity and reliability of smear restoration system.

After image blurring recovery, in order to evaluate restoration algorithm effect, needing to carry out quality evaluation to restored image, thus guaranteeing the restored image obtaining Quality advance.Owing to there is no picture rich in detail as with reference to image, non-reference picture quality appraisement can only be carried out.Current non-reference picture quality appraisement mainly contains two class methods, one is the gradient information of simple computed image, as gradient magnitude, border width, edge sharpness etc., this kind of index is not owing to considering the factor such as brightness of image, contrast, evaluation result is more unilateral, is particularly difficult to identify to the ringing effect of restored image; Two is adopt study mechanism, by a large amount of sample training, draw evaluation result, because sample can not comprise all situations, the performance of this method depends on the large size of sample, once the less situation that will occur data poor fitting of sample, if require higher to nicety of grading in training, there will be the over-fitting of data, no matter occur that situation, picture appraisal result all can be caused inaccurate, after smear restoration process, Output rusults is unstable, can not be used for real system.

Chinese patent CN101996406A proposes a kind of index of structure based sharpness to carry out reference-free quality evaluation, this index improves in structural similarity index, structural similarity is a kind of full reference image quality appraisement index, weigh by calculating the similarity of information to two width images of two width images in brightness, contrast and degree of structuration three, if wherein piece image is picture rich in detail, then can draw the evaluation quality of another piece image, but this index does not consider gradient information, the quality assessment of blurred picture cannot be used for.This patent portrays the impact at edge by the structural similarity of compute gradient image, but due to three indexs gradient image all, what reflect is gradient image analog information, well can not evaluate the similarity of original image, also just can not get an objective appraisal result.

A kind of typical non-reference picture quality appraisement index is average gray gradient G MG (Gray Mean Gradients), this method is by calculating the average gradient value of gray level image as the standard weighing picture quality, this method parameter is single, do not consider the information such as the brightness of image, contrast, do not consider the correlativity between pixel yet.In addition, when there is artificial trace in image as situations such as ringing effects, the average gradient of image can be caused to become large, and the fact is image quality decrease, also just can not get an objective appraisal result, GMG index can not process this kind of situation about often running in smear restoration.

Summary of the invention

It is poor to the object of the invention is to solve current existing method real-time performance, effectively can not differentiate blurred picture and picture rich in detail, effective evaluation can not be carried out to image restoration result, the problem of Output rusults instability after smear restoration process, and the discriminating of a kind of fast digital image blur and restored image method for evaluating quality are proposed.

Above-mentioned goal of the invention is achieved through the following technical solutions:

Step one: input gray level blurred picture F (i, j), obtains the size M × N of gray scale blurred picture F (i, j), calculates the horizontal gradient image G of gray scale blurred picture F (i, j) _x(i, j) and VG (vertical gradient) image and G _y(i, j), thus obtain gradient image G (i, j); Wherein, M is the width of image, and N is the height of image;

The average gray Grad GMG of step 2: computed image F (i, j);

Step 3: the histogram h of compute gradient image G (i, j), by traveling through histogram h, obtains the statistical distribution information NGN of gradient image;

Step 4: carry out product calculation according to GMG and NGN that step 2 and step 3 obtain, obtains comprehensive fuzzy identification beacon BIM; Threshold determination is carried out to the BIM obtained, if BIM is greater than predetermined threshold value T, is then judged to be that picture rich in detail directly exports; If BIM is less than predetermined threshold value T, then F (i, j) is proceeded to step 5; Wherein, threshold value T obtains according to picture rich in detail and blurred picture statistical study;

Step 5: adopt typical blindly restoring image algorithm to carry out recovery operation to gray scale blurred picture F (i, j), obtain restored image D (i, j);

Step 6: use size is m × m, variance is σ ²gaussian Blur check the reference picture D1 that gray scale blurred picture F (i, j) and restored image D (i, j) carry out reference picture F1 and D (i, j) of secondary fuzzy generation F (i, j); Wherein, the fuzzy use Gaussian Blur of secondary, m is the size of fuzzy core;

Step 7: according to the brightness similarity of F (i, j), F1, D (i, j) and D1 computing reference image and image to be evaluated, contrast similarity and structural similarity; Wherein, image to be evaluated is F (i, j) and D (i, j); Reference picture is F1 and D1; The brightness similarity of reference picture and image to be evaluated comprises l (F (i, j), F1) and l (D (i, j), D1); The contrast similarity c (F (i, j), F1) of reference picture and image to be evaluated and c (D (i, j), D1); The structural similarity s (F (i, j), F1) of reference picture and image to be evaluated and s (D (i, j), D1); L (F (i, j), F1) be two width image F (i, j) and the brightness similarity of F1, l (D (i, j), D1) be two width image D (i, and brightness similarity, the c (F (i of D1 j), j), F1) be the contrast similarity of two width images F (i, j) and F1; C (D (i, j), D1) is the contrast similarity of two width images D (i, j) and D1; S (F (i, j), F1) is the structural similarity of two width images F (i, j) and F1; S (D (i, j), D1) is the structural similarity of two width images D (i, j) and D1;

Step 8: the gradient image g of computing reference image and image to be evaluated respectively _f, g _f1, g _d, g _d1, utilize gradient image to calculate gradient similarity g (F (i, j), F1) and the g (D (i, j), D1) of two width images respectively; Wherein, g _ffor the gradient image of F (i, j), g _f1for the gradient image of F1, g _dfor the gradient image of D (i, j); g _d1for the gradient image of D1; G (F (i, j), F1) is two width images F (i, j) and F1 gradient similarity; G (D (i, j), D1) is the gradient similarity of two width images D (i, j) and D1;

Step 9: the structural similarity index ISSIM improved according to the brightness similarity of reference picture and image to be evaluated, contrast similarity, structural similarity and gradient Similarity Measure, the structural similarity index ISSIM piecemeal of improvement is tried to achieve M ISSIM index to get its mean value again and obtain MISSIM, calculate non-reference picture evaluation index INRSS according to MISSIM; Wherein, the MISSIM obtained specifically comprises MISSIM (F (i, j), F1) and MISSIM (D (i, j), D1); The structural similarity index ISSIM improved comprises two width image F (i, and the gradient index of similarity ISSIM (F (i of F1 j), j), F1) and two width image D (i, and the gradient index of similarity ISSIM (D (i of D1 j), j), D1);

The computing formula of the structural similarity index ISSIM of described improvement is:

ISSIM(F(i,j),F1)＝[l(F(i,j),F1)] ^α[c(F(i,j),F1)] ^β[s(F(i,j),F1)] ^γ[g(F(i,j),F1)] ^λ；

ISSIM(D(i,j),D1)＝[l(D(i,j),D1)] ^α[c(D(i,j),D1)] ^β[s(D(i,j),D1)] ^γ[g(D(i,j),D1)] ^λ

Wherein, α, beta, gamma, λ is every weight;

Step 10: judge according to the non-reference picture evaluation index INRSS that step 9 obtains, INRSS1>INRSS2 then illustrates that quality is not improved, then smear restoration system output gray level blurred picture F (i, j), otherwise key diagram picture element amount improves and exports restored image D (i, j); Wherein, $\begin{array}{c}\mathrm{INRSS}1=1-\mathrm{MISSIM}(F(i,j),F1)\\ \mathrm{INRSS}2=1-\mathrm{MISSIM}(D(i,j),D1)\end{array};$ Namely complete a kind of fast digital image blur to differentiate and restored image method for evaluating quality.

Invention effect

The present invention is a kind of improvement for current blur image restoration technology, both can be used for fuzzy discriminating and the restored image quality assessment work of single image, and also can be used for real time video processing, carry out effective video recuperation more.

The present invention be directed to gray scale blurred picture, for coloured image, only the component extraction of each passage out need to be processed respectively, finally again index be averaged; For existing digital fuzzy video and image restoration system Problems existing, introduce fuzzy identification module and fuzzy discriminating is carried out to pending image, on the PC platform of typical 4GB internal memory, 2.5GHz processor, fuzzy discrimination process is consuming time lower than 1ms, if spectral discrimination is picture rich in detail, then can be used for the image procossing of relevant subsequent the time of each frame 15 ~ 20ms in military domain, for associated video processes the condition of providing in real time.Propose a set of image quality measure method to image structure information and image detail and the comprehensive assessment of marginal information energy simultaneously and quality assessment is carried out to the image after smear restoration, avoid the artificial trace introduced by Image Restoration Algorithm to cause restored image quality to be inferior to problem that original image still exports to user or video and next link of pattern information processing system, particularly carried image quality measure method effectively can identify the adverse effect such as ringing effect and edge fog factor, thus the target recognition and tracking system based on image is effectively worked there is important value.

And for the deficiency that existing smear restoration system exists, this patent mainly solves following technical matters:

1) for current digital imagery smear restoration system, not there is fuzzy distinguishing ability to cause carrying out deblurring restoration disposal to picture rich in detail yet, thus make the problem of restored image quality degradation and waste messaging resource, fuzzy identification module is added to blur image restoration system, propose effective fuzzy identification algorithm, make it to identify the picture frame needing to carry out smear restoration, add the realization that fuzzy identification module is conducive to real time video processing, there is very strong practicality; As Fig. 2 (a), Fig. 2 (b), Fig. 3 (a) and Fig. 3 (b);

2) adopt the statistical law of nature picture rich in detail prior distribution to carry out fuzzy identification algorithm design, fuzzy identification algorithm has scale invariability simultaneously, has the scope of application widely, and this algorithm computation process is simple in addition, and exploitativeness is strong;

3) problem of restored image deterioration is caused for the artificial trace introduced in blur image restoration process, propose a kind of non-reference picture method for evaluating quality of novel improved, thus can assess accurately and effectively restored image quality, avoid restoring output image quality than the ropy problem of initial input blurred picture as Fig. 2 (a) and Fig. 2 (b); Ensure that the reliability of image restoration system, if carry out restoration disposal to a width picture rich in detail, then likely cause Fig. 3 (a) and Fig. 3 (b), shown effect, picture quality clearly after process is lower than quality before treatment, owing to having obvious ringing effect in the image after process, GMG etc. can draw higher evaluation of estimate based on the evaluation index of single gradient information, cause the Output rusults obtaining mistake.And adopt index that the present invention puies forward, owing to considering the correlativity between image pixel, index result can obtain one and tend to fuzzy value, because the correlativity between the larger local neighbor of gradient is less, the structural similarity obtained is less, identifies ringing effect by this species diversity.

4) the image quality measure method carried is by introducing the structure of image, edge and interior details information, strengthen the accuracy of restored image quality evaluation, good distinguishing ability is had especially to the ringing effect that restored image exists, ensure that the objectivity of image quality evaluation and INRSS's of the present invention (the non-reference picture quality appraisement index of the structure based similarity of improvement) is changed to-37.8%, compared with GMG be changed to-15.0% rate of change larger.Make the reliability that this invention ensures that image restoration system;

5) can only for there being picture rich in detail situation as a reference for full reference image quality appraisement method, the present invention, by improving for non-reference picture quality appraisement to full reference image quality appraisement method, has actual applicability widely.

The present invention can differentiate ringing effect mainly for GMG index, because evaluation index considers the structural information of image, gradient information just portrays edge as a part for index, instead of using gradient as sole indicator.The structural information of image is portrayed mainly through the correlativity calculated between pixel, and the correlativity of the local pixel that gradient is larger is less, so just can offset the impact of a part of ring on evaluation result, play certain discriminating effect.

The invention provides a kind of smear restoration system evaluated based on fuzzy discriminating and recovering quality, belong to image and field of video processing.Specifically, the present invention can be used for improving digital imagery smear restoration system work efficiency, improve digital imagery smear restoration system information processing rate and digital image blur recovery effect is effectively assessed.Can be applicable to the fields such as the shooting of civilian Digital photographic, intelligent video monitoring, remote sensing of optical imaging, military imaging reconnaissance and missile imaging guidance.

Fuzzy discriminating step is introduced in digital image blurring recovery system by the present invention first, saves the processing time greatly, for real time video processing task provides more computational resources compared to conventional digital blur image restoration system.Meanwhile, in order to ensure the robustness of restoration result, the present invention has introduced the quality assessment step of restored image, makes output image quality be not less than input picture, ensure that the reliability of smear restoration system.In image blurring discriminating, invention introduces image gradient amplitude and gradient and to distribute two aspect indexs, feature the feature of picture rich in detail more accurately, it is more accurate to differentiate.In restored image quality assessment, the image quality evaluating method of existing structure based similarity is improved, add the gradient index of similarity of picture engraving edge and detailed information, the adverse effect factor such as can evaluate ringing effect in restored image better and blur margin is clear, reflects restored image quality more objectively.Use this patent can significantly improve real-time and the reliability of digital fuzzy video and image restoration system, the application request developing the more satisfied relevant dual-use system of corresponding system made.

Accompanying drawing explanation

A kind of fast digital image blur of Fig. 1 is differentiated and restored image method for evaluating quality process flow diagram;

Fig. 2 (a) is gray scale blurred picture;

Fig. 2 (b) is the improved image after smear restoration quality assessment;

Fig. 3 (a) differentiates, for picture rich in detail, not need the picture rich in detail carrying out Fuzzy Processing through fuzzy identification module;

Fig. 3 (b), for differentiate without fuzzy identification module, restores the image of the Quality Down obtained to picture rich in detail through classic method;

Fig. 4 is the non-reference picture quality appraisement index INRSS calculation flow chart improved.

embodiment

Embodiment one: a kind of fast digital image blur of present embodiment is differentiated and restored image method for evaluating quality, specifically prepares according to following steps:

Step one: input gray level blurred picture F (i, j), obtains the size M × N of gray scale blurred picture F (i, j), calculates the horizontal gradient image G of gray scale blurred picture F (i, j) _x(i, j) and VG (vertical gradient) image and G _y(i, j), thus obtain gradient image G (i, j); Wherein, M is the width of image, and N is the height of image;

The average gray Grad GMG (Gray Mean Gradients) of step 2: computed image F (i, j), this index reflects the gradient magnitude characteristic of the fuzzy identification beacon of image;

Step 3: statistics with histogram is carried out to gradient image, i.e. compute gradient image G (i, j) histogram h, by traveling through histogram h, obtain the non-zero gray shade scale number NGN (Non-zero Gray Gradient Number) that histogram of gradients h is corresponding, namely obtain the statistical distribution information NGN of gradient image;

Step 4: carry out product calculation (BIM=NGN × GMG) according to GMG and NGN that step 2 and step 3 obtain, obtains comprehensive fuzzy identification beacon BIM (Blur Identification Metric); Threshold determination is carried out to the BIM obtained, if BIM is greater than predetermined threshold value T, is then judged to be that picture rich in detail directly exports; If BIM is less than predetermined threshold value T, then F (i, j) is proceeded to step 5; Wherein, threshold value T obtains according to picture rich in detail and blurred picture statistical study, namely by carrying out analysis integrated to different scene image, draw Fuzzy Threshold T, also threshold value can be changed according to patient fog-level in actual applications, make this index have good adaptability, after definite threshold, can fuzzy judgement be carried out;

Step 5: adopt typical blindly restoring image algorithm to carry out recovery operation as methods such as Tikhonov regularization, full variational regularization to gray scale blurred picture F (i, j), obtain restored image D (i, j);

Step 6: use size is m × m, variance is σ ²gaussian Blur check the reference picture D1 that gray scale blurred picture F (i, j) and restored image D (i, j) carry out reference picture F1 and D (i, j) of secondary fuzzy generation F (i, j); Wherein, the fuzzy use Gaussian Blur of secondary, m is the size of fuzzy core;

Step 7: according to the brightness similarity of F (i, j), F1, D (i, j) and D1 computing reference image and image to be evaluated, contrast similarity and structural similarity; Wherein, image to be evaluated is F (i, j) and D (i, j); Reference picture is F1 and D1; The brightness similarity of reference picture and image to be evaluated comprises l (F (i, j), F1) and l (D (i, j), D1); The contrast similarity c (F (i, j), F1) of reference picture and image to be evaluated and c (D (i, j), D1); The structural similarity s (F (i, j), F1) of reference picture and image to be evaluated and s (D (i, j), D1); L (F (i, j), F1) be two width image F (i, j) and the brightness similarity of F1, l (D (i, j), D1) be two width image D (i, and brightness similarity, the c (F (i of D1 j), j), F1) be the contrast similarity of two width images F (i, j) and F1; C (D (i, j), D1) is the contrast similarity of two width images D (i, j) and D1; S (F (i, j), F1) is the structural similarity of two width images F (i, j) and F1; S (D (i, j), D1) is the structural similarity of two width images D (i, j) and D1;

Step 8: the gradient image g of computing reference image and image to be evaluated respectively _f, g _f1, g _d, g _d1, utilize gradient image to calculate gradient similarity g (F (i, j), F1) and the g (D (i, j), D1) of two width images respectively; Wherein, g _ffor the gradient image of F (i, j), g _f1for the gradient image of F1, g _dfor the gradient image of D (i, j); g _d1for the gradient image of D1; G (F (i, j), F1) is two width images F (i, j) and F1 gradient similarity; G (D (i, j), D1) is the gradient similarity of two width images D (i, j) and D1;

Step 9: according to the brightness similarity of reference picture and image to be evaluated, contrast similarity, the structural similarity index ISSIM (Improved Structural Similarity) that structural similarity and gradient Similarity Measure improve, for reducing local difference, the structural similarity index ISSIM piecemeal of improvement is tried to achieve M ISSIM index to get its mean value again and obtain MISSIM (Mean of Improved Structural Similarity), non-reference picture evaluation index INRSS (Improved No-Reference Structural Similarity) is calculated according to MISSIM, namely the comprehensive evaluation result of blurred picture and restored image quality comprises brightness, contrast, degree of structuration and marginal information, step 6 ~ nine process flow diagram is as Fig. 4, wherein, the MISSIM obtained specifically comprises MISSIM (F (i, j), F1) and MISSIM (D (i, j), D1), the structural similarity index ISSIM improved comprises two width image F (i, and the gradient index of similarity ISSIM (F (i of F1 j), j), F1) and two width image D (i, and the gradient index of similarity ISSIM (D (i of D1 j), j), D1),

The computing formula of the structural similarity index ISSIM of described improvement is:

ISSIM(F(i,j),F1)＝[l(F(i,j),F1)] ^α[c(F(i,j),F1)] ^β[s(F(i,j),F1)] ^γ[g(F(i,j),F1)] ^λ；

ISSIM(D(i,j),D1)＝[l(D(i,j),D1)] ^α[c(D(i,j),D1)] ^β[s(D(i,j),D1)] ^γ[g(D(i,j),D1)] ^λ

Wherein, α, beta, gamma, λ is every weight;

Step 10: judge according to the non-reference picture evaluation index INRSS that step 9 obtains, INRSS1>INRSS2 then illustrates that quality is not improved, for ensureing the stability of restoration result, then smear restoration system output gray level blurred picture F (i, j), output restored image D (i, j) otherwise key diagram picture element amount improves, wherein $\begin{array}{c}\mathrm{INRSS}1=1-\mathrm{MISSIM}(F(i,j),F1)\\ \mathrm{INRSS}2=1-\mathrm{MISSIM}(D(i,j),D1)\end{array};$ Differentiate and restored image method for evaluating quality as namely Fig. 1 completes a kind of fast digital image blur.

Embodiment two: present embodiment and embodiment one unlike: obtain gradient image G (i, j) in step one, computing formula is as follows:

$\begin{array}{c}{G}_x(i,j)=|F(i+1,j)-F(i,j)|\\ G_y(i,j)=|F(i,j+1)-F(i,j)|\\ G(i,j)=\left[\begin{array}{cc}{G}_x(i,j)& G_y(i,j)\end{array}\right]\end{array},$ Wherein, i, j are respectively horizontal stroke corresponding to image slices vegetarian refreshments, ordinate value.Other step and parameter identical with embodiment one.

Embodiment three: present embodiment and embodiment one or two unlike: in step 2, the average gray Grad GMG formula of computed image F (i, j) is as follows:

$\mathrm{GMG}=\frac{\underset{i=1}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\sqrt{\frac{{[F(i+1,j)-F(i,j)]}^2+{[F(i,j+1)-F(i,j)]}^2}{2}}}{(M-1)(N-1)}$

Wherein, M gray scale blurred picture F (i, j) length, the width of N gray scale blurred picture F (i, j).Other step and parameter identical with embodiment one or two.

Embodiment four: one of present embodiment and embodiment one to three unlike: obtain the non-zero gray shade scale number NGN in h in step 3, computing formula is as follows:

h＝hist(G(i,j))

NGN＝num(h _i≠0)

In formula, hist represents histogram manipulation; h _ifor the pixel count that i-th gray level in h is corresponding, by calculating NGN index, the trailing portion in picture engraving hangover distribution, in order to differentiate the fog-level of image.Other step and parameter identical with one of embodiment one to three.

Embodiment five: one of present embodiment and embodiment one to four unlike: to gray scale blurred picture F (i in step 5, j) typical blindly restoring image algorithm and full variational regularization method is adopted to carry out blind recovery, can certainly adopt other blind restoration methods, its rejuvenation target function is as follows:

$\underset{f}{\arg \min }\{\frac{\mathrm{\λ}}{2}{\left|\right|k\⊗f-F(i,j)\left|\right|}^2+{\left|\right|G_x\left|\right|}_1+{\left|\right|G_y\left|\right|}_1\}$

Wherein, k is unknown fuzzy core, and f is unknown picture rich in detail, for convolution algorithm, for adjustable parameter, G _xfor horizontal gradient image, G _yfor VG (vertical gradient) image, get final product final restored image D (i, j) by alternately solving above-mentioned rejuvenation target function; Blurred picture and good recovery effect as shown in Figure 2 a and 2 b, with obvious artificial trace poor recovery effect as shown in Figure 3 a and Figure 3 b shows; Fig. 3 a and Fig. 3 b gives prominence to the importance of fuzzy discriminating, and follow-up is the equal of other one heavily to ensure to the evaluation of this two width image, even if guarantee to have carried out smear restoration operation to picture rich in detail mistakenly, also by quality assessment identify, the better picture rich in detail of output quality.Other step and parameter identical with one of embodiment one to four.

Embodiment six: one of present embodiment and embodiment one to five unlike: the detailed process calculating the brightness similarity of reference picture and image to be evaluated, contrast similarity and structural similarity in step 7 is:

Respectively to F (i, j), the first three items of the structural similarity of F1, D (i, j) and D1 computed improved, i.e. brightness similarity l (F (i, j), F1), l (D (i, j), D1), contrast similarity c (F (i, j), F1), c (D (i, j), and structural similarity s (F (i, j), F1) D1), s (D (i, j), D1), computing formula is as follows:

$l(F(i,j),F1)=\frac{2{\mathrm{\μ}}_F{\mathrm{\μ}}_{F1}+C_1}{{\mathrm{\μ}}_F^2+{\mathrm{\μ}}_{F1}^2+C_1},l(D(i,j),D1)=\frac{2{\mathrm{\μ}}_D{\mathrm{\μ}}_{D1}+C_1}{{\mathrm{\μ}}_D^2+{\mathrm{\μ}}_{D1}^2+C_1}$

$c(F(i,j),F1)=\frac{2{\mathrm{\σ}}_F{\mathrm{\σ}}_{F1}+C_2}{{\mathrm{\σ}}_F^2+{\mathrm{\σ}}_{F1}^2+C_2},c(D(i,j),D1)=\frac{2{\mathrm{\σ}}_D{\mathrm{\σ}}_{D1}+C_2}{{\mathrm{\σ}}_D^2+{\mathrm{\σ}}_{D1}^2+C_2}$

$s(F(i,j),F1)=\frac{2{\mathrm{\σ}}_{\mathrm{FF}1}+C_2}{{\mathrm{\σ}}_F{\mathrm{\σ}}_{F1}+C_2},s(D(i,j),D1)=\frac{2{\mathrm{\σ}}_{\mathrm{DD}1}{+C}_2}{{\mathrm{\σ}}_D{\mathrm{\σ}}_{D1}+C_2}$

Wherein, μ _ffor the average of F (i, j) image intensity value, μ _f1for the average of F1 image intensity value, μ _dfor the average of D (i, j) image intensity value, μ _d1for the average of D1 image intensity value, σ _ffor F (i, j) image intensity value variance, σ _f1for F1 image intensity value variance, σ _dfor D (i, j) image intensity value variance, σ _d1for D1 image intensity value variance, σ _fF1for the covariance of the gray-scale value of F (i, j) and F1 image, σ _dD1for the covariance of D (i, j) image and D1 image intensity value, C ₁=(k1 × L) ², C ₂=(k2 × L) ², k1=0.01, k2=0.03, L are pixel grayscale.Other step and parameter identical with one of embodiment one to five.

Embodiment seven: one of present embodiment and embodiment one to six are unlike the gradient image g of respectively computing reference image and image to be evaluated in step 8 _f, g _f1, g _d, g _d1, utilize gradient image to calculate the gradient similarity g (F (i, j), F1) of two width images respectively and g (D (i, j), D1) detailed process is:

Respectively to F (i, j), F1 and D (i, j), D1 compute gradient image g _f, g _f1, g _d, g _d1, then compute gradient image g _f, g _f1, g _d, g _d1gradient similarity, computing formula is as follows:

$g(F(i,j),F1)=\frac{2\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{g}_F(i,j)g_{F1}(i,j)+C_4}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}[{\left(g_F(i,j)\right)}^2+{\left(g_{F1}(i,j)\right)}^2]+C_4}$

$g(D(i,j),D1)=\frac{2\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{g}_D(i,j)g_{D1}(i,j)+C_4}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}[{\left(g_D(i,j)\right)}^2+{\left(g_{D1}(i,j)\right)}^2]+C_4}$

Wherein, C ₄=(k4 × L) ², k4=0.03, L are pixel grayscale, C ₄be a constant, guarantee that denominator is non-vanishing; In the marginal information of image also being taken into account that index of similarity is evaluated by this step, the quality of blurred picture can be evaluated better; Simultaneously compared to other sharpness evaluation indexes, can identify better smear restoration effect poor time the ringing effect introduced, thus ensure the stability of recovery system.Other step and parameter identical with one of embodiment one to six.

Embodiment eight: one of present embodiment and embodiment one to seven unlike: calculate non-reference picture evaluation index INRSS correlation computations formula according to MISSIM in step 9 as follows:

(1) regulate according to actual needs; When actual computation, M block is divided into by image to obtain its ISSIM index respectively; Try to achieve mean value MSSIM, namely adopt following formula to calculate: $\begin{array}{c}\mathrm{MTSSIM}(F(i,j),F1)=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{ISSIM}(F_i,{F1}_i)\\ \mathrm{MISSIM}(D(i,j),D1)=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{ISSIM}(D_i,{D1}_i)\end{array}$

F _ifor i-th subimage in gray scale blurred picture F (i, j), D _ii-th subimage in image D (i, j) after recovery; F1 _ifor i-th subimage in reference picture F1; D1 _ii-th subimage in reference picture D1;

(2) obtain the nothing reference evaluation index INRSS improved, computing formula is as follows:

$\begin{array}{c}\mathrm{INRSS}1=1-\mathrm{MISSIM}(F(i,j),F1)\\ \mathrm{INRSS}2=1-\mathrm{MISSIM}(D(i,j),D1)\end{array}.$ Other step and parameter identical with one of embodiment one to seven.

Following examples are adopted to verify beneficial effect of the present invention:

Embodiment one:

A kind of fast digital image blur of the present embodiment is differentiated and restored image method for evaluating quality, specifically prepares according to following steps:

Step one: input gray level blurred picture F (i, j), obtains the size M × N of gray scale blurred picture F (i, j), calculates the horizontal gradient image G of gray scale blurred picture F (i, j) _x(i, j) and VG (vertical gradient) image and G _y(i, j), thus obtain gradient image G (i, j); Wherein, M is the width of image, and N is the height of image;

Obtain gradient image G (i, j), computing formula is as follows:

$\begin{array}{c}{G}_x(i,j)=|F(i+1,j)-F(i,j)|\\ G_y(i,j)=|F(i,j+1)-F(i,j)|\\ G(i,j)=\left[\begin{array}{cc}{G}_x(i,j)& G_y(i,j)\end{array}\right]\end{array},$ Wherein, i, j are respectively horizontal stroke corresponding to image slices vegetarian refreshments, ordinate value;

The average gray Grad GMG (Gray Mean Gradients) of step 2: computed image F (i, j), this index reflects the gradient magnitude characteristic of the fuzzy identification beacon of image;

The average gray Grad GMG formula of computed image F (i, j) is as follows:

$\mathrm{GMG}=\frac{\underset{i=1}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\sqrt{\frac{{[F(i+1,j)-F(i,j)]}^2+{[F(i,j+1)-F(i,j)]}^2}{2}}}{(M-1)(N-1)}$

Wherein, M gray scale blurred picture F (i, j) length, the width of N gray scale blurred picture F (i, j).

Step 3: statistics with histogram is carried out to gradient image, i.e. compute gradient image G (i, j) histogram h, by traveling through histogram h, obtain the non-zero gray shade scale number NGN (Non-zero Gray Gradient Number) that histogram of gradients h is corresponding, namely obtain the statistical distribution information NGN of gradient image;

Obtain the non-zero gray shade scale number NGN in h, computing formula is as follows:

h＝hist(G(i,j))

NGN＝num(h _i≠0)

In formula, hist represents histogram manipulation; h _ifor the pixel count that i-th gray level in h is corresponding, by calculating NGN index, the trailing portion in picture engraving hangover distribution, in order to differentiate the fog-level of image.

Step 4: carry out product calculation (BIM=NGN × GMG) according to GMG and NGN that step 2 and step 3 obtain, obtains comprehensive fuzzy identification beacon BIM (Blur Identification Metric); Threshold determination is carried out to the BIM obtained, if BIM is greater than predetermined threshold value T, is then judged to be that picture rich in detail directly exports; If BIM is less than predetermined threshold value T, then F (i, j) is proceeded to step 5; Wherein, threshold value T obtains according to picture rich in detail and blurred picture statistical study, namely by carrying out analysis integrated to different scene image, draw Fuzzy Threshold T, also threshold value can be changed according to patient fog-level in actual applications, make this index have good adaptability, can carry out fuzzy judgement after definite threshold, general value is 700 ~ 900;

Step 5: adopt typical blindly restoring image algorithm and full variational regularization method to carry out blind recovery to gray scale blurred picture F (i, j), can certainly adopt other blind restoration methods, its rejuvenation target function is as follows:

$\underset{f}{\arg \min }\{\frac{\mathrm{\λ}}{2}{\left|\right|k\⊗f-F(i,j)\left|\right|}^2+{\left|\right|G_x\left|\right|}_1+{\left|\right|G_y\left|\right|}_1\}$

Wherein, k is unknown fuzzy core, and f is unknown picture rich in detail, for convolution algorithm, for adjustable parameter, G _xfor horizontal gradient image, G _yfor VG (vertical gradient) image, get final product final restored image D (i, j) by alternately solving above-mentioned rejuvenation target function; Blurred picture and good recovery effect as shown in Figure 2 a and 2 b, with obvious artificial trace poor recovery effect as shown in Figure 3 a and Figure 3 b shows.Fig. 3 a and Fig. 3 b gives prominence to the importance of fuzzy discriminating, and follow-up is the equal of other one heavily to ensure to the evaluation of this two width image, even if guarantee to have carried out smear restoration operation to picture rich in detail mistakenly, also by quality assessment identify, the better picture rich in detail of output quality.

Step 6: use size is m × m, variance is σ ²gaussian Blur check the reference picture D1 that gray scale blurred picture F (i, j) and restored image D (i, j) carry out reference picture F1 and D (i, j) of secondary fuzzy generation F (i, j); Here use m × m size to be 7 × 7, variance be 6 Gaussian Blur core carry out secondary Fuzzy Processing, be ensure secondary blur effect, m and σ ²general value is respectively 5 ~ 8; Wherein, the fuzzy use Gaussian Blur of secondary, m is the size of fuzzy core;

Step 7: be according to the brightness similarity of F (i, j), F1, D (i, j) and D1 computing reference image and image to be evaluated, contrast similarity and structural similarity detailed process:

Respectively to F (i, j), the first three items of the structural similarity of F1, D (i, j) and D1 computed improved, i.e. brightness similarity l (F (i, j), F1), l (D (i, j), D1), contrast similarity c (F (i, j), F1), c (D (i, j), and structural similarity s (F (i, j), F1) D1), s (D (i, j), D1), computing formula is as follows:

$l(F(i,j),F1)=\frac{2{\mathrm{\μ}}_F{\mathrm{\μ}}_{F1}+C_1}{{\mathrm{\μ}}_F^2+{\mathrm{\μ}}_{F1}^2+C_1},l(D(i,j),D1)=\frac{2{\mathrm{\μ}}_D{\mathrm{\μ}}_{D1}+C_1}{{\mathrm{\μ}}_D^2+{\mathrm{\μ}}_{D1}^2+C_1}$

$c(F(i,j),F1)=\frac{2{\mathrm{\σ}}_F{\mathrm{\σ}}_{F1}+C_2}{{\mathrm{\σ}}_F^2+{\mathrm{\σ}}_{F1}^2+C_2},c(D(i,j),D1)=\frac{2{\mathrm{\σ}}_D{\mathrm{\σ}}_{D1}+C_2}{{\mathrm{\σ}}_D^2+{\mathrm{\σ}}_{D1}^2+C_2}$

$s(F(i,j),F1)=\frac{2{\mathrm{\σ}}_{\mathrm{FF}1}+C_2}{{\mathrm{\σ}}_F{\mathrm{\σ}}_{F1}+C_2},s(D(i,j),D1)=\frac{2{\mathrm{\σ}}_{\mathrm{DD}1}{+C}_2}{{\mathrm{\σ}}_D{\mathrm{\σ}}_{D1}+C_2}$

Wherein, μ _ffor the average of F (i, j) image intensity value, μ _f1for the average of F1 image intensity value, μ _dfor the average of D (i, j) image intensity value, μ _d1for the average of D1 image intensity value, σ _ffor F (i, j) image intensity value variance, σ _f1for F1 image intensity value variance, σ _dfor D (i, j) image intensity value variance, σ _d1for D1 image intensity value variance, σ _fF1for the covariance of the gray-scale value of F (i, j) and F1 image, σ _dD1for the covariance of D (i, j) image and D1 image intensity value, C ₁=(k1 × L) ², C ₂=(k2 × L) ², k1=0.01, k2=0.03, L are pixel grayscale, are generally 255.Image to be evaluated is F (i, j) and D (i, j); Reference picture is F1 and D1; The brightness similarity of reference picture and image to be evaluated comprises l (F (i, j), F1) and l (D (i, j), D1); The contrast similarity c (F (i, j), F1) of reference picture and image to be evaluated and c (D (i, j), D1); The structural similarity s (F (i, j), F1) of reference picture and image to be evaluated and s (D (i, j), D1); L (F (i, j), F1) be two width image F (i, j) and the brightness similarity of F1, l (D (i, j), D1) be two width image D (i, and brightness similarity, the c (F (i of D1 j), j), F1) be the contrast similarity of two width images F (i, j) and F1; C (D (i, j), D1) is the contrast similarity of two width images D (i, j) and D1; S (F (i, j), F1) is the structural similarity of two width images F (i, j) and F1; S (D (i, j), D1) is the structural similarity of two width images D (i, j) and D1;

Step 8: the gradient image g of computing reference image and image to be evaluated respectively _f, g _f1, g _d, g _d1, utilize gradient image to calculate gradient similarity g (F (i, j), F1) and the g (D (i, j), D1) of two width images respectively;

Respectively to F (i, j), F1 and D (i, j), D1 compute gradient image g _f, g _f1, g _d, g _d1, then compute gradient image g _f, g _f1, g _d, g _d1gradient similarity, computing formula is as follows:

$g(F(i,j),F1)=\frac{2\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{g}_F(i,j)g_{F1}(i,j)+C_4}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}[{\left(g_F(i,j)\right)}^2+{\left(g_{F1}(i,j)\right)}^2]+C_4}$

$g(D(i,j),D1)=\frac{2\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{g}_D(i,j)g_{D1}(i,j)+C_4}{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}[{\left(g_D(i,j)\right)}^2+{\left(g_{D1}(i,j)\right)}^2]+C_4}$

Wherein, C ₄=(k4 × L) ², k4=0.03, L are pixel grayscale, are generally 255, C ₄be a constant, guarantee that denominator is non-vanishing.In the marginal information of image also being taken into account that index of similarity is evaluated by this step, the quality of blurred picture can be evaluated better.Simultaneously compared to other sharpness evaluation indexes, can identify better smear restoration effect poor time the ringing effect introduced, thus ensure the stability of recovery system.G _ffor the gradient image of F (i, j), g _f1for the gradient image of F1, g _dfor the gradient image of D (i, j); g _d1for the gradient image of D1; G (F (i, j), F1) is two width images F (i, j) and F1 gradient similarity; G (D (i, j), D1) is the gradient similarity of two width images D (i, j) and D1;

Step 9: according to the brightness similarity of reference picture and image to be evaluated, contrast similarity, the structural similarity index ISSIM (Improved Structural Similarity) that structural similarity and gradient Similarity Measure improve, for reducing local difference, the structural similarity index ISSIM piecemeal of improvement is tried to achieve M ISSIM index to get its mean value again and obtain MISSIM (Mean of Improved Structural Similarity), non-reference picture evaluation index INRSS (Improved No-Reference Structural Similarity) is calculated according to MISSIM, namely the comprehensive evaluation result of blurred picture and restored image quality comprises brightness, contrast, degree of structuration and marginal information, M ∈ (1,12) step 6 ~ nine process flow diagram is as Fig. 4, wherein, the MISSIM obtained specifically comprises MISSIM (F (i, j), F1) and MISSIM (D (i, j), D1), the structural similarity index ISSIM improved comprises two width image F (i, and the gradient index of similarity ISSIM (F (i of F1 j), j), F1) and two width image D (i, and the gradient index of similarity ISSIM (D (i of D1 j), j), D1),

The computing formula of the structural similarity index ISSIM of described improvement is:

ISSIM(F(i,j),F1)＝[l(F(i,j),F1)] ^α[c(F(i,j),F1)] ^β[s(F(i,j),F1)] ^γ[g(F(i,j),F1)] ^λ；

ISSIM(D(i,j),D1)＝[l(D(i,j),D1)] ^α[c(D(i,j),D1)] ^β[s(D(i,j),D1)] ^γ[g(D(i,j),D1)] ^λ

Wherein, α, beta, gamma, λ is every weight;

(1) regulate according to actual needs; When actual computation, M block is divided into by image to obtain its ISSIM index respectively; Try to achieve mean value MSSIM, namely adopt following formula to calculate: $\begin{array}{c}\mathrm{MTSSIM}(F(i,j),F1)=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{ISSIM}(F_i,{F1}_i)\\ \mathrm{MISSIM}(D(i,j),D1)=\frac{1}{M}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\mathrm{ISSIM}(D_i,{D1}_i)\end{array}$

F _ifor i-th subimage in gray scale blurred picture F (i, j), D _ii-th subimage in image D (i, j) after recovery; F1 _ifor i-th subimage in reference picture F1; D1 _ii-th subimage in reference picture D1;

(2) obtain the nothing reference evaluation index INRSS improved, computing formula is as follows:

INRSS1＝1-MISSIM(F(i,j),F1)

INRSS2＝1-MISSIM(D(i,j),D1)

The INRSS evaluation result obtained Fig. 2 a and Fig. 2 b two width image is as shown in the table:

Index	Before recovery	After recovery	Change
				INRSS	0.3172	0.5769	29.05％

INRSS index is obtained to Fig. 3 (a) and Fig. 3 (b) two width image and GMG index comparing result is as shown in the table:

Index	Before recovery	After recovery	Change
				INRSS	0.4063	0.2527	-37.8％
GMG	7.0012	5.9559	-15.0％

Result shows that INRSS has good evaluation effect to image, and more effectively can distinguish the impact of artificial trace on restoration result.

Step 10: judge according to the non-reference picture evaluation index INRSS that step 9 obtains, INRSS1>INRSS2 then illustrates that quality is not improved, for ensureing the stability of restoration result, then smear restoration system output gray level blurred picture F (i, j), output restored image D (i, j) otherwise key diagram picture element amount improves, wherein $\begin{array}{c}\mathrm{INRSS}1=1-\mathrm{MISSIM}(F(i,j),F1)\\ \mathrm{INRSS}2=1-\mathrm{MISSIM}(D(i,j),D1)\end{array};$ Differentiate and restored image method for evaluating quality as namely Fig. 1 completes a kind of fast digital image blur.

The present invention also can have other various embodiments; when not deviating from the present invention's spirit and essence thereof; those skilled in the art are when making various corresponding change and distortion according to the present invention, but these change accordingly and are out of shape the protection domain that all should belong to the claim appended by the present invention.

Claims (8)

1. fast digital image blur is differentiated and a restored image method for evaluating quality, it is characterized in that: a kind of fast digital image blur discriminating and restored image method for evaluating quality specifically carry out according to the following steps:

Step one: input gray level blurred picture F (i, j), obtains the size M × N of gray scale blurred picture F (i, j), calculates the horizontal gradient image G of gray scale blurred picture F (i, j) _x(i, j) and VG (vertical gradient) image and G _y(i, j), thus obtain gradient image G (i, j); Wherein, M is the width of image, and N is the height of image;

The average gray Grad GMG of step 2: computed image F (i, j);

Step 3: the histogram h of compute gradient image G (i, j), by traveling through histogram h, obtains the statistical distribution information NGN of gradient image;

Step 4: carry out product calculation according to GMG and NGN that step 2 and step 3 obtain, obtains comprehensive fuzzy identification beacon BIM; Threshold determination is carried out to the BIM obtained, if BIM is greater than predetermined threshold value T, is then judged to be that picture rich in detail directly exports; If BIM is less than predetermined threshold value T, then F (i, j) is proceeded to step 5; Wherein, threshold value T obtains according to picture rich in detail and blurred picture statistical study;

Step 5: adopt typical blindly restoring image algorithm to carry out recovery operation to gray scale blurred picture F (i, j), obtain restored image D (i, j);

Step 6: use size is m × m, variance is σ ²gaussian Blur check the reference picture D1 that gray scale blurred picture F (i, j) and restored image D (i, j) carry out reference picture F1 and D (i, j) of secondary fuzzy generation F (i, j); Wherein, the fuzzy use Gaussian Blur of secondary, m is the size of fuzzy core;

Step 7: according to the brightness similarity of F (i, j), F1, D (i, j) and D1 computing reference image and image to be evaluated, contrast similarity and structural similarity; Wherein, image to be evaluated is F (i, j) and D (i, j); Reference picture is F1 and D1; The brightness similarity of reference picture and image to be evaluated comprises l (F (i, j), F1) and l (D (i, j), D1); The contrast similarity c (F (i, j), F1) of reference picture and image to be evaluated and c (D (i, j), D1); The structural similarity s (F (i, j), F1) of reference picture and image to be evaluated and s (D (i, j), D1); L (F (i, j), F1) be two width image F (i, j) and the brightness similarity of F1, l (D (i, j), D1) be two width image D (i, and brightness similarity, the c (F (i of D1 j), j), F1) be the contrast similarity of two width images F (i, j) and F1; C (D (i, j), D1) is the contrast similarity of two width images D (i, j) and D1; S (F (i, j), F1) is the structural similarity of two width images F (i, j) and F1; S (D (i, j), D1) is the structural similarity of two width images D (i, j) and D1;

Step 8: the gradient image g of computing reference image and image to be evaluated respectively _f, g _f1, g _d, g _d1, utilize gradient image to calculate gradient similarity g (F (i, j), F1) and the g (D (i, j), D1) of two width images respectively; Wherein, g _ffor the gradient image of F (i, j), g _f1for the gradient image of F1, g _dfor the gradient image of D (i, j); g _d1for the gradient image of D1; G (F (i, j), F1) is two width images F (i, j) and F1 gradient similarity; G (D (i, j), D1) is the gradient similarity of two width images D (i, j) and D1;

Step 9: the structural similarity index ISSIM improved according to the brightness similarity of reference picture and image to be evaluated, contrast similarity, structural similarity and gradient Similarity Measure, the structural similarity index ISSIM piecemeal of improvement is tried to achieve M ISSIM index to get its mean value again and obtain MISSIM, calculate non-reference picture evaluation index INRSS according to MISSIM; Wherein, the MISSIM obtained specifically comprises MISSIM (F (i, j), F1) and MISSIM (D (i, j), D1); The structural similarity index ISSIM improved comprises two width image F (i, and the gradient index of similarity ISSIM (F (i of F1 j), j), F1) and two width image D (i, and the gradient index of similarity ISSIM (D (i of D1 j), j), D1); The computing formula of the structural similarity index ISSIM of described improvement is:

ISSIM(F(i,j),F1)＝[l(F(i,j),F1)] ^α[c(F(i,j),F1)] ^β[s(F(i,j),F1)] ^γ[g(F(i,j),F1)] ^λ

；

ISSIM(D(i,j),D1)＝[l(D(i,j),D1)] ^α[c(D(i,j),D1)] ^β[s(D(i,j),D1)] ^γ[g(D(i,j),D1)] ^λ

Wherein, α, beta, gamma, λ is every weight;

Step 10: judge according to the non-reference picture evaluation index INRSS that step 9 obtains, INRSS1>INRSS2 then illustrates that quality is not improved, then smear restoration system output gray level blurred picture F (i, j), otherwise key diagram picture element amount improves and exports restored image D (i, j); Wherein, namely complete a kind of fast digital image blur to differentiate and restored image method for evaluating quality.

2. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: obtain gradient image G (i, j) in step one, computing formula is as follows:

wherein, i, j are respectively horizontal stroke corresponding to image slices vegetarian refreshments, ordinate value.

3. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: in step 2, the average gray Grad GMG formula of computed image F (i, j) is as follows:

Wherein, M gray scale blurred picture F (i, j) length, the width of N gray scale blurred picture F (i, j).

4. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: obtain the non-zero gray shade scale number NGN in h in step 3, computing formula is as follows:

h＝hist(G(i,j))

NGN＝num(h _i≠0)

In formula, hist represents histogram manipulation; h _ifor the pixel count that i-th gray level in h is corresponding.

5. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: to gray scale blurred picture F (i in step 5, j) adopt typical blindly restoring image algorithm to adopt full variational regularization method to carry out blind recovery, its rejuvenation target function is as follows:

Wherein, k is unknown fuzzy core, and f is unknown picture rich in detail, for convolution algorithm, λ, for adjustable parameter, G _xfor horizontal gradient image, G _yfor VG (vertical gradient) image, get final product final restored image D (i, j) by alternately solving above-mentioned rejuvenation target function.

6. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: the detailed process calculating the brightness similarity of reference picture and image to be evaluated, contrast similarity and structural similarity in step 7 is:

Respectively brightness similarity l (F (i, j) is calculated to F (i, j), F1, D (i, j) and D1, F1), l (D (i, j), D1), contrast similarity c (F (i, j), F1), c (D (i, j), D1) and structural similarity s (F (i, j), F1), s (D (i, j), D1), computing formula is as follows:

Wherein, μ _ffor the average of F (i, j) image intensity value, μ _f1for the average of F1 image intensity value, μ _dfor the average of D (i, j) image intensity value, μ _d1for the average of D1 image intensity value, σ _ffor F (i, j) image intensity value variance, σ _f1for F1 image intensity value variance, σ _dfor D (i, j) image intensity value variance, σ _d1for D1 image intensity value variance, σ _fF1for the covariance of the gray-scale value of F (i, j) and F1 image, σ _dD1for the covariance of D (i, j) image and D1 image intensity value, C ₁=(k1 × L) ², C ₂=(k2 × L) ², k1=0.01, k2=0.03, L are pixel grayscale.

7. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: the gradient image g of difference computing reference image and image to be evaluated in step 8 _f, g _f1, g _d, g _d1, utilize gradient image to calculate the gradient similarity g (F (i, j), F1) of two width images respectively and g (D (i, j), D1) detailed process is:

Respectively to F (i, j), F1 and D (i, j), D1 compute gradient image g _f, g _f1, g _d, g _d1, then compute gradient image g _f, g _f1, g _d, g _d1gradient similarity, computing formula is as follows:

Wherein, C ₄=(k4 × L) ², k4=0.03, L are pixel grayscale, C ₄it is a constant.

8. a kind of fast digital image blur is differentiated and restored image method for evaluating quality according to claim 1, it is characterized in that: calculate non-reference picture evaluation index INRSS correlation computations formula according to MISSIM in step 9 as follows:

(1) M block is divided into by image to obtain its ISSIM index respectively; Try to achieve mean value MSSIM, namely adopt following formula to calculate:

F _ifor i-th subimage in gray scale blurred picture F (i, j), D _ii-th subimage in image D (i, j) after recovery; F1 _ifor i-th subimage in reference picture F1; D1 _ii-th subimage in reference picture D1;

(2) obtain the nothing reference evaluation index INRSS improved, computing formula is as follows:

INRSS1＝1-MISSIM(F(i,j),F1)

。

INRSS2＝1-MISSIM(D(i,j),D1)。