CN104376565A - Non-reference image quality evaluation method based on discrete cosine transform and sparse representation - Google Patents

Abstract

The invention discloses a non-reference image quality evaluation method based on discrete cosine transform and sparse representation to mainly solve the problem that in the prior art, non-reference image quality evaluation is not accurate. The method comprises the following steps that a gray level image is input, discrete cosine transform is carried out on the gray level image, and natural scene statistical characteristics are extracted; natural scene statistical characteristics of a series of images of different distortion types and different content are extracted, and an original characteristic dictionary is established according to the average subjective difference score; clustering is carried out on the original characteristics dictionary, and atoms are selected in a self-adaptation mode according to the tested image characteristics and the approximation degrees in the original characteristic dictionary to form a sparse representation dictionary; the tested image characteristics are solved and the sparse representation coefficients are calculated through sparse representation in characteristics space, linear weighting summation is carried out according to the subjective evaluation values in the sparse representation dictionary, and the image quality measure is obtained. The method has good consistency with the subjective evaluation result and is suitable for quality evaluation on images with various distortion types.

Description

Based on the non-reference picture quality appraisement method of discrete cosine transform and rarefaction representation

Technical field

The invention belongs to image processing field, relate to the objective evaluation of picture quality, can be used for image acquisition, compression coding, Internet Transmission.

Background technology

Image is the important channel of mankind's obtaining information, and picture quality represents that image provides the ability of information to people or equipment, and direct relation adequacy and the accuracy of institute's obtaining information.But image is in the process obtaining, process, transmit and store, and because various factors impact inevitably will produce degradation problems, this brings extreme difficulties to the acquisition of information or the post-processed of image.Therefore, effective image quality evaluation mechanism is set up extremely important.As can be used for Performance comparision, the Selecting parameter of various algorithm in the processing procedure such as image denoising, image co-registration; Can be used for instructing the transmitting procedure of whole image and assessment of system performance at Image Coding and the communications field; In addition, be also significant in scientific domains such as image processing algorithm optimization, living things feature recognitions.

Image quality evaluation can be divided into subjective evaluation method and method for objectively evaluating method.Subjective method relies on the subjective perception of experimenter to carry out assess image quality, and objective method is then according to the quantizating index that model provides, and simulating human visual system perceives mechanism weighs picture quality.Due to the final recipient that people is image, therefore subjective quality assessment is the most reliable evaluation method.Subjective method the most frequently used is at present mean subjective percentile method MOS and difference subjective score method DMOS, but be subject to the impact of subjective factor due to its evaluation result, consuming time and can not automatically realize when amount of images is large, therefore the research of objective image evaluation method is just seemed particular importance.

According to the degree of dependence to reference picture when evaluating, Objective image quality evaluation can be divided into full reference image quality appraisement FR-IQA, partial reference image quality appraisement RR-IQA and non-reference picture quality appraisement NR-IQA.

The great advantage of full reference image quality appraisement FR-IQA method is accurate to distorted image prediction of quality, and full reference method conventional at present has the method MSE and PSNR, the method SSIM of structure based similarity and the method HVS etc. based on human visual system that add up based on pixel error.But all need the priori that original image is complete due to these methods, need to store and the data volume of transmission comparatively large, limit its application in many practical field, therefore partial reference type image quality evaluating method becomes one of focus of people's research.

Partial reference image quality appraisement RR-IQA method does not need complete original reference image, but needs to utilize the Feature Combination of some reference pictures to obtain the massfraction of distorted image.Although evaluation method of can ensuring the quality of products on the basis reducing required transmission information amount possesses good accuracy, but still need transmit the partial information of original image.In most of practical application, the information of original image cannot obtain at all or procurement cost very high.

It is a kind of any prior imformation not needing original image that no-reference image quality evaluates NR-IQA method, directly distorted image is carried out to the method for quality evaluation.Because people understand limited to human visual system and corresponding brain cognitive process at present, the Design and implementation of its algorithm is more difficult.Current non-reference picture quality appraisement method has: " the Z.M.P.Sazzad; Y.Kawayoke; and Y.Horita; No-reference image quality assessment for jpeg2000based on spatial features; Signal Process.Image Commun.; vol.23 of the non-reference picture quality appraisement method for JP2000 distortion that Sazzad etc. propose, no.4, pp.257 – 268, Apr.2008 ", but the method is not only suitable for for JPEG2000 compression and evaluates the impact on image such as fuzzy, noise.Moorthy etc. propose method " the A.K.Moorthyand A.C.Bovik; A two-step framework for constructing blind image quality indices; IEEE SignalProcess.Lett.; vol.17; no.5; pp.513 – 516 based on study, May 2010. "; this method is the mapping relations of direct modeling characteristics of image and quality; its mathematical model and model parameter all need training or manually obtain; because mapping relations can not accurate analog, cause final mass evaluation result not accurate enough.Anush Krishna Moorthy etc. also been proposed the method " the Distortion Identification-based Image Verity and INtegrity Evaluation (DIIVINE) index " based on natural scene statistics, the method evaluation result and subjective assessment have the consistance of height, but 88 dimensions are reached to the feature that a sub-picture extracts, characteristic extraction procedure is consuming time long, and huge characteristic dimension is also inconvenient to classify and the training of regression model.

Summary of the invention

The object of the invention is to for above-mentioned existing methodical deficiency, propose a kind of non-reference picture quality appraisement method based on discrete cosine transform and rarefaction representation, extracted data message can be made full use of while reduction image zooming-out characteristic dimension, by the method for rarefaction representation to image feature information High Efficiency Modeling, improve the accuracy of quality evaluation result.

Realize the object of the invention technical scheme to comprise the steps:

(1) read in a width gray level image I, discrete cosine transform is carried out to it, extract a series of natural scene statistical nature f be associated with subjective perception;

(2) the primitive character dictionary of training image is set up;

(2a) repeat step (1), extract the natural scene statistical nature of the secondary training image of n, and constitutive characteristic matrix F, F=[f ₁, f ₂..., f _i..., f _n], wherein f _ibe the natural scene statistical nature of the i-th secondary training image, i=1,2 ..., n;

(2b) integrate the mean subjective discrepancy score of the secondary training image of n, and form quality vector M, M=[m ₁, m ₂..., m _i..., m _n], wherein m _ibe the mean subjective discrepancy score of the i-th secondary training image, i=1,2 ..., n;

(2c) quality vector M and eigenmatrix F is carried out correspondence to combine, builds primitive character dictionary D:

$D=\left[\begin{array}{c}F\\ M\end{array}\right]=[{\stackrel{\‾}{f}}_1,{\stackrel{\‾}{f}}_2,...,{\stackrel{\‾}{f}}_i,...,{\stackrel{\‾}{f}}_n],$ Wherein ${\stackrel{\‾}{f}}_i=\left[\begin{array}{c}{f}_i\\ m_i\end{array}\right],i=\mathrm{1,2},...,n;$

(3) gather for H class with K-means algorithm by primitive character dictionary D, the cluster centre of kth class is C _k, wherein $C_k=\left[\begin{array}{c}{\mathrm{Fc}}_k\\ {\mathrm{Mc}}_k\end{array}\right],$ Fc _kfor feature clustering center, Mc _kfor clustering quality center, k=1,2 ..., H;

(4) read in a secondary test pattern I', repeat step (1), extract the natural scene statistical nature f' of test pattern I';

(5) the feature clustering center Fc of kth class in the natural scene statistical nature f' of test pattern I' and primitive character dictionary D is calculated _keuclidean distance dis _k, use P _krepresent the degree of approximation of kth class in this test pattern and primitive character dictionary D, wherein

$P_k=\frac{1/{\mathrm{dis}}_k}{{\mathrm{\Σ}}_{k=1}^{H}1/{\mathrm{dis}}_k},k=\mathrm{1,2},...,H;$

(6) according to degree of approximation P _k, select and feature clustering center Fc from the kth class of primitive character dictionary D _kthe minimum front N of Euclidean distance _kindividual sample is as the rarefaction representation atom of test pattern I' wherein i=1,2 ..., N, k=1,2 ..., H, H class atom forms the rarefaction representation dictionary D' for test pattern I' jointly:

$D^{\′}=[D_1,D_2,...,D_k,...,D_H]=\left[\begin{array}{c}{F}^{\′}\\ M^{\′}\end{array}\right],$ Wherein $D_k=[{\stackrel{\‾}{f}}_1^k,{\stackrel{\‾}{f}}_2^k,...,{\stackrel{\‾}{f}}_i^k,...,{\stackrel{\‾}{f}}_{N_k}^k],{\stackrel{\‾}{f}}_i^k=\left[\begin{array}{c}{f}_i^k\\ m_i^k\end{array}\right],$ F _i ^kfor the natural scene statistical nature of i-th sample in the kth class selected from primitive character dictionary D, for the mean subjective discrepancy score of the sample of i-th in kth class, i=1,2 ..., N _k, N _k=δ * P _k, k=1,2 ..., H, δ are normal number;

(7) the rarefaction representation coefficient of natural scene statistical nature f' under eigenmatrix F' of test pattern I' is solved according to the method for rarefaction representation: wherein argmin represents and will make objective function λ || α || ₁+ || f'-D ' α || ₂minimum α=[α ¹, α ²..., α ^k..., α ^h] ^tassignment is to α ^*, k=1,2 ..., H, i=1,2 ..., N _k, R ^lrepresent that L ties up real number space, represent vectorial β=[β ₁, β ₂..., β _l..., β _l] ^t1 norm, represent 2 norms of vectorial β, l=1,2 ..., L, λ are the normal numbers for balancing fidelity item and regular terms, and T represents matrix transpose operation;

(8) quality of test pattern I' is calculated according to constructed rarefaction representation: q ∈ [0,100], wherein for kth class i-th atom f in rarefaction representation dictionary D' _i ^kmassfraction, α _k,irepresent natural scene statistical nature f' kth class i-th atom f in eigenmatrix F' of test pattern I' _i ^kexpression coefficient, Q is that the final mass of test pattern I' is estimated;

The present invention compared with prior art has the following advantages:

1. the present invention can make full use of the information of database, carries out dictionary training study, thus can carry out quality evaluation to the image of different type of distortion to the natural scene statistical nature of type of distortion image different in database.With existing great majority only for certain distortion type non-reference picture quality appraisement algorithm compared with, the scope of application is wider.

2. the present invention utilizes the characteristics dictionary trained, without any need for original reference image information, directly carry out Efficient Evaluation to picture quality, compare more convenient with partial reference type image quality evaluating method with full-reference image quality evaluating method, range of application is wider.

3. the present invention to fc-specific test FC image can in the characteristics of image dictionary of having trained the corresponding rarefaction representation dictionary of adaptive selection, compared with other existing algorithms, evaluation result and subjective consistency better.

Accompanying drawing explanation

Fig. 1 is realization flow figure of the present invention;

Fig. 2 is the natural scene statistical nature extracted after carrying out discrete cosine transform to gray level image I in the present invention;

Fig. 3 is the result to primitive character dictionary D K-means cluster in the present invention;

Fig. 4 is to the objective assessment score of test pattern and mean subjective discrepancy score comparison diagram with the present invention.

Embodiment

Below in conjunction with accompanying drawing, specific embodiment of the invention step and effect are described in further detail:

With reference to Fig. 1, performing step of the present invention is as follows:

Step 1, extracts the natural scene statistical nature f of gray level image I on first yardstick _s;

(1a) read in a secondary gray level image I, what gray level image I is decomposed into 5*5 has overlapping image block each other, carries out discrete cosine transform, and remove the DC component of discrete cosine transform coefficient to each image block;

(1b) respectively Generalized Gaussian Distribution Model matching is used to the discrete cosine transform coefficient after the removal DC component of each image block, obtain the form factor γ of each image block, get the average of the form factor γ of all image blocks as the natural scene statistical nature f on first yardstick _{s, 1}first element f _{s, 1,1}, to the form factor of all image blocks according to arranging from small to large, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}second element f _{s, 1,2}, the Generalized Gaussian Distribution Model function wherein calculating form factor γ is: $f\left(x\right|\mathrm{\α},\mathrm{\β},\mathrm{\γ})=\mathrm{\α}*e^{-{\left(\mathrm{\β}{(x-\mathrm{\μ})}^T{\mathrm{\Σ}}^{-1}(x-\mathrm{\μ})\right)}^{\mathrm{\γ}}},\mathrm{\β}=\frac{1}{\mathrm{\σ}}\sqrt{\frac{\mathrm{\Γ}(3/\mathrm{\γ})}{\mathrm{\Γ}(1/\mathrm{\γ})}},$ μ is average, and σ is variance, and γ is form factor, and β is scale factor;

(1c) the frequency change coefficient of each image block is calculated get the frequency change coefficient of all image blocks average as the natural scene statistical nature f on first yardstick _{s, 1}the 3rd element f _{s, 1,3}, to the frequency change coefficient of all image blocks according to arranging from big to small, get the average of front 10% as the natural scene statistical nature f on first yardstick _s, ₁the 4th element f _{s, Isosorbide-5-Nitrae}, wherein μ _{| x|}and σ _{| x|}be respectively average and the standard deviation of the absolute value of the discrete cosine transform coefficient of the removal DC component of each image block, || represent and ask absolute value;

(1d) according to the order from high frequency to low frequency, the discrete cosine transform coefficient of each image block is divided into 3 frequency bands, calculates the frequency band energy rate of change R of each image block _n, be specifically calculated as follows formula:

wherein represent the variance of the n-th frequency band, n represents each frequency band, E _nrepresent the energy of the n-th frequency band, n=1,2,3, j is positive integer;

(1e) get all image blocks frequency band energy rate of change R _naverage as the natural scene statistical nature f on first yardstick _{s, 1}the 5th element f _{s, 1,5}, to the frequency band energy rate of change R of all image blocks _naccording to arranging from big to small, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}the 6th element f _{s, 1,6};

(1f) discrete cosine transform coefficient of each image block is divided into along 45 °, 90 °, the subband in 135 ° of these 3 directions, and with generalized Gaussian distribution function respectively to the subband matching in each direction, obtain the frequency change coefficient of 3 directional subbands, and calculate the variance of the frequency change coefficient of these 3 directional subbands

(1g) variance of the frequency change coefficient of 3 directional subbands of all image blocks is got average as the natural scene statistical nature f on first yardstick _{s, 1}the 7th element f _{s, 1,7}, to the variance of frequency change coefficient of 3 directional subbands of getting all image blocks according to arranging from big to small, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}the 8th element f _{s, 1,8};

(1h) the 8 dimension natural scene statistical nature fs of gray level image I on first yardstick are obtained _{s, 1}:

f _s,1＝[f _s,1,1,f _s,1,2,…,f _s,1,i…,f _s,1,8]，i＝1,2,…,8。

Step 2, extracts the natural scene statistical nature f of gray level image I on multiple yardstick;

(2a) down-sampling operation is carried out to gray level image I and obtain down-sampled images I2, step 1 is repeated to down-sampled images I2 and operates, obtain the natural scene statistical nature f of gray level image I on second yardstick _{s, 2};

(2b) down-sampling operation is carried out again to down-sampled images I2 and obtain secondary down-sampled images I3, step 1 is repeated to secondary down-sampled images I3 and operates, obtain the natural scene statistical nature f of gray level image I on the 3rd yardstick _{s, 3};

(2c) the natural scene statistical nature of gray level image I is tried to achieve: f=[f _{s, 1}, f _{s, 2}, f _{s, 3}] ^t, T represents matrix transpose operation.

Step 3, extracts the natural scene statistical nature f' of test pattern I' according to step 1 and step 2;

Step 4, builds primitive character dictionary D;

(4a) in training set, successively step 1 is carried out to every width image and step 2 operates, extract the natural scene statistical nature of the secondary training image of n, and constitutive characteristic matrix F, F=[f ₁, f ₂..., f _i..., f _n], wherein f _ibe the natural scene statistical nature of the i-th secondary training image, i=1,2 ..., n;

(4b) integrate the mean subjective discrepancy score of the secondary training image of n, and form quality vector M, M=[m ₁, m ₂..., m _i..., m _n], wherein m _ibe the mean subjective discrepancy score of the i-th secondary training image, i=1,2 ..., n;

(4c) quality vector M and eigenmatrix F is carried out correspondence to combine, builds primitive character dictionary D: $D=\left[\begin{array}{c}F\\ M\end{array}\right]=[{\stackrel{\‾}{f}}_1,{\stackrel{\‾}{f}}_2,...,{\stackrel{\‾}{f}}_i,...,{\stackrel{\‾}{f}}_n],$ Wherein ${\stackrel{\‾}{f}}_i=\left[\begin{array}{c}{f}_i\\ m_i\end{array}\right],i=\mathrm{1,2},...,n;$

Step 5, selects corresponding rarefaction representation dictionary for specific test pattern in characteristics of image dictionary;

(5a) gather for H class with K-means algorithm by primitive character dictionary D, in H class, kth class cluster centre is C _k, wherein $C_k=\left[\begin{array}{c}{\mathrm{Fc}}_k\\ {\mathrm{Mc}}_k\end{array}\right],$ Fc _kfor feature clustering center, Mc _kfor clustering quality center, k=1,2 ... H;

(5b) read in a secondary test pattern I', carry out step 1 successively and step 2 operates, extract the natural scene statistical nature f' of test pattern I';

(5c) the feature clustering center Fc of kth class in the natural scene statistical nature f' of test pattern I' and primitive character dictionary D is calculated _keuclidean distance dis _k, obtain the degree of approximation P of kth class in test pattern I' and primitive character dictionary D _k, wherein

$P_k=\frac{1/{\mathrm{dis}}_k}{{\mathrm{\Σ}}_{k=1}^{H}1/{\mathrm{dis}}_k},k=\mathrm{1,2},...,H;$

(5d) according to degree of approximation P _k, select and feature clustering center Fc from the kth class of primitive character dictionary D _kthe minimum front N of Euclidean distance _kindividual sample is as the rarefaction representation atom of test pattern I' wherein i=1,2 ..., N, k=1,2 ..., H, H class atom forms the rarefaction representation dictionary D' for test pattern I' jointly:

$D^{\′}=[D_1,D_2,...,D_k,...,D_H]=\left[\begin{array}{c}{F}^{\′}\\ M^{\′}\end{array}\right],$ Wherein $D_k=[{\stackrel{\‾}{f}}_1^k,{\stackrel{\‾}{f}}_2^k,...,{\stackrel{\‾}{f}}_i^k,...,{\stackrel{\‾}{f}}_{N_k}^k],{\stackrel{\‾}{f}}_i^k=\left[\begin{array}{c}{f}_i^k\\ m_i^k\end{array}\right],$ F _i ^kfor in the kth class selected from primitive character dictionary D _ithe natural scene statistical nature of individual sample, for the mean subjective discrepancy score of the sample of i-th in kth class, i=1,2 ..., N _k, N _k=δ * P _k, k=1,2 ..., H, δ are normal number.

Step 6, the feature of rarefaction representation test pattern;

The rarefaction representation coefficient of natural scene statistical nature f' under eigenmatrix F' of test pattern I' is solved according to the method for rarefaction representation: wherein arg min represents and will make objective function λ || α || ₁+ || f'-D ' α || ₂minimum α=[α ¹, α ²..., α ^k..., α ^h] ^tassignment is to α ^*, k=1,2 ..., H, i=1,2 ..., N _k, R ^lrepresent that L ties up real number space, represent vectorial β=[β ₁, β ₂..., β _l..., β _l] ^t1 norm, represent 2 norms of vectorial β, λ is the normal number for balancing fidelity item and regular terms, and T represents matrix transpose operation.

Step 7, the quality assessment utilizing following formula to calculate test pattern I' is estimated:

q ∈ [0,100], wherein for kth class i-th atom f in rarefaction representation dictionary D' _i ^kmassfraction, α _k,irepresent natural scene statistical nature f' kth class i-th atom f in eigenmatrix F' of test pattern I' _i ^kexpression coefficient, Q is that the final mass of test pattern I' is estimated, and the quality of Q value less explanation test pattern is higher.

Effect of the present invention can be further illustrated by following experiment:

1. experiment condition and standards of grading:

This test carries out on the second generation LIVE image quality measure database of TEXAS university of the U.S., this database comprises the distorted image of 29 high-resolution undistorted RGB color image and five corresponding types, comprises the image of 175 width jpeg images, 169 width JPEG2000 images, 145 width white noise WN images, 145 width Gaussian Blur Gblur images and the distortion after fast-fading FF channel of 145 width.The mean subjective discrepancy score that database gives distorted image describes the quality of distorted image.

In order to test the consistance of picture quality objective evaluation result and subjective perception that the present invention proposes, the following 3 kinds of modules of this experimental selection: is linearly dependent coefficient LCC, reflect the accuracy that method for objectively evaluating is predicted; Two is root-mean-square error RMSE, reflects the error of method for objectively evaluating; Three is Spearman rank order correlation coefficient S ROCC, reflects the monotonicity of objective evaluation prediction of result.

In test, image data base is divided into training plan image set and test pattern image set, and training plan image set is for training primitive character dictionary, and test pattern is used for prediction and evaluation result.The packet mode of employing two kinds is had according to the difference of original image number.First group: training set has 15 width original images; Second group: training set has 23 width original images.

Experiment parameter is arranged: 1) during primitive character dictionary cluster, H gets 4; 2), when forming rarefaction representation dictionary, δ gets 4; 3), when solving rarefaction representation coefficient with rarefaction representation, λ rule of thumb gets 0.0001.

2. experiment content and result:

The evaluation of experiment 1:LIVE_database2 database epigraph distortion

The non-reference picture quality appraisement method based on discrete cosine transform and rarefaction representation utilizing the present invention to propose, use two kinds of packet modes to concentrate the test pattern of LIVE_database2 database and carry out quality evaluation, computed image Objective Quality Assessment result and the conforming 3 kinds of modules of subjective perception respectively: linearly dependent coefficient LCC, Spearman rank order correlation coefficient S ROCC and root-mean-square error RMSE.

Table 1 gives the test and appraisal of the distortion on LIVE_database2 database.Wherein DCTSR1 is the test result that the present invention uses the first packet mode, DCTSR2 is the test result that the present invention uses the second packet mode, wavelet transformation and rarefaction representation associated methods 1 are the test result that wavelet transformation and rarefaction representation associated methods use the first packet mode, and wavelet transformation and rarefaction representation associated methods 2 are the test result that wavelet transformation and rarefaction representation associated methods use the second packet mode.

The comparative test result of table 1 this method and other image quality evaluating methods

As can be seen from Table 1, when using same packets mode, the present invention most ofly has good superiority with existing compared with reference method: 1) have higher forecasting accuracy, namely linearly dependent coefficient LCC is larger than existing most methods; 2) have the monotonicity of stricter prediction, namely rank order correlation coefficient S ROCC is larger than existing most methods; 3) the present invention can predict very well to the image of various type of distortion.

Experiment 2: picture quality objective evaluation result and subjective perception consistance are tested

The quality assessment of the test pattern experimentally obtained in 1 is estimated, draw the loose some graph of a relation that the mean subjective discrepancy score DMOS of Q and test pattern is estimated in the quality assessment of the present invention to test pattern, and use logarithmic function matching to obtain optimum matching curve, as shown in Figure 4.In Fig. 4, the quality assessment of horizontal ordinate representative image is estimated, the mean subjective discrepancy score DMOS of ordinate representative image, ' ' represents the image quality evaluation that the present invention obtains and estimates, and the distribution of ' ' is more close with the optimum matching curve simulated, and illustrates that effect is better.Wherein:

4 (a) is the effect of the present invention in JPEG2000 compression artefacts image word bank,

4 (b) is the effect of the present invention in JPEG compression artefacts image word bank,

4 (c) is the effect of the present invention in white noise WN distorted image word bank,

4 (d) is the effect of the present invention in Gaussian Blur Gblur distorted image word bank,

4 (e) is the effect of the present invention in the image word bank of distortion after fast-fading FF channel,

4 (f) is the effect of the present invention in whole LIVE_database2 database.

As can be seen from Figure 4, relatively, deflection curve is less, and in the different distorted image word banks of whole database, performance is all very stable, higher with subjective vision perception consistance for the distribution of the present invention's ' ' and the optimum matching curve simulated.

Claims (2)

1., based on a non-reference picture quality appraisement method for discrete cosine transform and rarefaction representation, comprise the steps:

(1) read in a width gray level image I, discrete cosine transform is carried out to it, extract a series of natural scene statistical nature f be associated with subjective perception;

(2) the primitive character dictionary of training image is set up;

(2a) repeat step (1), extract the natural scene statistical nature of the secondary training image of n, and constitutive characteristic matrix F, F=[f ₁, f ₂..., f _i..., f _n], wherein f _ibe the natural scene statistical nature of the n-th secondary training image, i=1,2 ..., n;

(2b) integrate the mean subjective discrepancy score of the secondary training image of n, and form quality vector M, M=[m ₁, m ₂..., m _i..., m _n], wherein m _ibe the mean subjective discrepancy score of the n-th secondary training image, i=1,2 ..., n;

(2c) quality vector M and eigenmatrix F is carried out correspondence to combine, builds primitive character dictionary D: $D=\left[\begin{array}{c}F\\ M\end{array}\right]=[\stackrel{\‾}{f_1},\stackrel{\‾}{f_2},...,\stackrel{\‾}{f_i},...,\stackrel{\‾}{f_n}],$ Wherein $\stackrel{\‾}{f_i}=\left[\begin{array}{c}{f}_i\\ m_i\end{array}\right],i=\mathrm{1,2},...,n;$

(2d) gather for H class with K-means algorithm by primitive character dictionary D, in H class, kth class cluster centre is C _k, wherein $C_k=\left[\begin{array}{c}{\mathrm{Fc}}_k\\ {\mathrm{Mc}}_k\end{array}\right],$ Fc _kfor feature clustering center, Mc _kfor clustering quality center, k=1,2...H.

(3) read in a secondary test pattern I', repeat step (1), extract the natural scene statistical nature f' of test pattern I';

(4) the feature clustering center Fc of kth class in the natural scene statistical nature f' of test pattern I' and primitive character dictionary D is calculated _keuclidean distance dis _k, obtain the degree of approximation P of kth class in test pattern I' and primitive character dictionary D _k, wherein $P_k=\frac{1/{\mathrm{dis}}_k}{{\mathrm{\Σ}}_{k=1}^{H}1/{\mathrm{dis}}_k},$ k＝1,2,...,H；

(5) according to degree of approximation P _k, select and feature clustering center Fc from the kth class of primitive character dictionary D _kthe minimum front N of Euclidean distance _kindividual sample is as the rarefaction representation atom of test pattern I' wherein i=1,2 ..., N, k=1,2 ..., H, H class atom forms the rarefaction representation dictionary D' for test pattern I' jointly: $D^{\′}=[D_1,D_2,...,D_k,...,D_H]=\left[\begin{array}{c}{F}^{\′}\\ M^{\′}\end{array}\right],$ Wherein $D_k=[{\stackrel{\‾}{f}}_1^k,{\stackrel{\‾}{f}}_2^k,...,{\stackrel{\‾}{f}}_i^k,...,{\stackrel{\‾}{f}}_{N_k}^k],{\stackrel{\‾}{f}}_i^k=\left[\begin{array}{c}{f}_i^k\\ m_i^k\end{array}\right],$ for the natural scene statistical nature of i-th sample in the kth class selected from primitive character dictionary D, for the mean subjective discrepancy score of the sample of i-th in kth class, i=1,2 ..., N _k, N _k=δ * P _k, k=1,2 ..., H, δ are normal number;

(6) the rarefaction representation coefficient of natural scene statistical nature f' under eigenmatrix F' of test pattern I' is solved according to the method for rarefaction representation: wherein argmin represents and will make objective function λ || α || ₁+ || f'-D ' α || ₂minimum α=[α ¹, α ²..., α ^k..., α ^h] ^tassignment is to α ^*, k=1,2 ..., H, i=1,2 ..., N _k, R ^lrepresent that L ties up real number space, $L=\underset{k=1}{\overset{H}{\mathrm{\Σ}}}{N}_k,{\left|\right|\mathrm{\β}\left|\right|}_1=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}\left|{\mathrm{\β}}_l\right|$ Represent vectorial β=[β ₁, β ₂..., β _l..., β _l] ^t1 norm, ${\left|\right|\mathrm{\β}\left|\right|}_2=\sqrt{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{\mathrm{\β}}_l^2}$ Represent 2 norms of vectorial β, λ is the normal number for balancing fidelity item and regular terms, and T represents matrix transpose operation;

(7) quality of test pattern I' is calculated according to constructed rarefaction representation: q ∈ [0,100], wherein for kth class i-th atom in rarefaction representation dictionary D' massfraction, α _k,irepresent natural scene statistical nature f' kth class i-th atom in eigenmatrix F' of test pattern I' expression coefficient, Q is that the final mass of test pattern I' is estimated.

2. the non-reference picture quality appraisement method based on discrete cosine transform and rarefaction representation according to claim 1, wherein described in step (1), discrete cosine transform is carried out to gray level image I, extract a series of natural scene statistical nature f be associated with subjective perception, concrete steps are as follows:

(1.1) the natural scene statistical nature f of gray level image I on first yardstick is extracted _{s, 1};

(1.1a) what gray level image I is decomposed into 5*5 has overlapping image block each other, carries out discrete cosine transform, and remove the DC component of discrete cosine transform coefficient to each image block;

(1.1b) respectively Generalized Gaussian Distribution Model matching is used to the discrete cosine transform coefficient after the removal DC component of each image block, obtain the form factor γ of each image block, get the average of the form factor γ of all image blocks as the natural scene statistical nature f on first yardstick _{s, 1}first element f _{s, 1,1}, to the form factor of all image blocks according to arranging from small to large, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}second element f _{s, 1,2}, the Generalized Gaussian Distribution Model function wherein calculating form factor γ is: $f\left(x\right|\mathrm{\α},\mathrm{\β},\mathrm{\γ})=\mathrm{\α}*e^{{-\left(\mathrm{\β}{(x-\mathrm{\μ})}^T{\mathrm{\Σ}}^{-1}(x-\mathrm{\μ})\right)}^{\mathrm{\γ}}},\mathrm{\β}=\frac{1}{\mathrm{\σ}}\sqrt{\frac{\mathrm{\Γ}(3/\mathrm{\γ})}{\mathrm{\Γ}(1/\mathrm{\γ})}},\mathrm{\Γ}\left(z\right)={\∫}_0^{\∞}{t}^{z-1}{e}^{-t}\mathrm{dt},\mathrm{\α}=\frac{\mathrm{\β\γ}}{2\mathrm{\Γ}(1/\mathrm{\γ})},$ μ is average, and σ is variance, and γ is form factor, and β is scale factor;

(1.1c) the frequency change coefficient of each image block is calculated get the frequency change coefficient of all image blocks average as the natural scene statistical nature f on first yardstick _{s, 1}the 3rd element f _{s, 1,3}, to the frequency change coefficient of all image blocks according to arranging from big to small, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}the 4th element f _{s, Isosorbide-5-Nitrae}, wherein μ _{| x|}and σ _{| x|}be respectively average and the standard deviation of the discrete cosine transform coefficient of the removal DC component of each image block, || represent and ask absolute value;

(1.1d) according to the order from high frequency to low frequency, the discrete cosine transform coefficient of each image block is divided into 3 frequency bands, calculates the frequency band energy rate of change R of each image block _n, be specifically calculated as follows formula:

$R_n=\frac{|E_n-\frac{1}{n-1}{\mathrm{\&Sigma;}}_{j<n}{E}_j|}{E_n+\frac{1}{n-1}{\mathrm{\&Sigma;}}_{j<n}{E}_j},$ Wherein $E_n={\mathrm{\&sigma;}}_n^2,$ N=1,2,3, represent the variance of the n-th frequency band, n represents each frequency band, E _nrepresent the energy of the n-th frequency band, j is positive integer;

(1.1e) get all image blocks frequency band energy rate of change R _naverage as the natural scene statistical nature f on first yardstick _{s, 1}the 5th element f _{s, 1,5}, to the frequency band energy rate of change R of all image blocks _naccording to arranging from big to small, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}the 6th element f _{s, 1,6};

(1.1f) discrete cosine transform coefficient of each image block is divided into along 45 °, 90 °, the subband in 135 ° of these 3 directions, and with generalized Gaussian distribution function respectively to the subband matching in each direction, obtain the frequency change coefficient of 3 directional subbands, and calculate the variance of the frequency change coefficient of these 3 directional subbands

(1.1g) variance of the frequency change coefficient of 3 directional subbands of all image blocks is got average as the natural scene statistical nature f on first yardstick _{s, 1}the 7th element f _{s, 1,7}, to the variance of frequency change coefficient of 3 directional subbands of getting all image blocks according to order arrangement from big to small, get the average of front 10% as the natural scene statistical nature f on first yardstick _{s, 1}the 8th element f _{s, 1,8};

(1.1h) the 8 dimension natural scene statistical nature fs of gray level image I on first yardstick are obtained _{s, 1}:

f _s,1＝[f _s,1,1,f _s,1,2,...,f _s,1,i...,f _s,1,8]，i＝1,2,...,8。

(1.2) down-sampling operation is carried out to gray level image I and obtain down-sampled images I2, step (1.1) operation is repeated to down-sampled images I2, obtains the natural scene statistical nature f of gray level image I on second yardstick _{s, 2};

(1.3) down-sampling operation is carried out again to down-sampled images I2 and obtain secondary down-sampled images I3, step (1.1) operation is repeated to secondary down-sampled images I3, obtains the natural scene statistical nature f of gray level image I on the 3rd yardstick _{s, 3};

(1.4) the natural scene statistical nature of gray level image I is tried to achieve: f=[f _{s, 1}, f _{s, 2}, f _{s, 3}] ^t, T represents matrix transpose operation.