CN107977967A - A kind of non-reference picture quality appraisement method towards visual angle synthesis - Google Patents
A kind of non-reference picture quality appraisement method towards visual angle synthesis Download PDFInfo
- Publication number
- CN107977967A CN107977967A CN201711399720.XA CN201711399720A CN107977967A CN 107977967 A CN107977967 A CN 107977967A CN 201711399720 A CN201711399720 A CN 201711399720A CN 107977967 A CN107977967 A CN 107977967A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- dog
- msup
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Abstract
The present invention proposes a kind of non-reference picture quality appraisement method towards visual angle synthesis, and this method quantifies the distortion in composograph by analyzing the characteristics of two class distortions, design feature：The distortion brought first to the edge damage of image and the unnatural property of texture quantify, and extract individual features.Then feature is integrated using the method for machine learning, so as to train the Environmental Evaluation Model that the distortion that can be brought to whole building-up process is evaluated.The present invention overcomes two shortcomings of existing method：(1) existing method can only evaluate a kind of distortion in building-up process, and the two class distortions that this method can be in the whole building-up process of effective evaluation.(2) existing method is largely full reference method, i.e., they must could carry out quality evaluation in the case where providing original undistorted image to distorted image, and context of methods is no reference method, has and is more widely applied prospect.
Description
Technical field
The present invention relates to virtual perspective synthesis objective visual quality evaluation method, it is especially a kind of towards visual angle synthesis
Non-reference picture quality appraisement method.
Background technology
Visual angle synthesis is that new multi-view image is synthesized using texture image and depth image.Visual angle synthetic technology is regarding more
There is very extensive application [1] in the fields such as angle video, free-viewing angle TV.Quality evaluation to the visual angle of synthesis, which can quantify, to be commented
The quality of valency visual angle synthetic technology, can also be used to optimum synthesis technology.In addition, the quality evaluation to visual angle composograph is straight
Connecing influences the success or not of these applications.So the quality evaluation towards visual angle synthesis is of great significance.
Distortion in synthesis visual angle is broadly divided into two classes.The first kind is to obtain, and handles and transmit texture image and depth
The traditional distortion brought in image process；Second class is the drafting distortion that empty filling is brought in the drawing process of visual angle.
The existing quality evaluating method for visual angle composograph is designed both for wherein independent a kind of distortion.For
A kind of distortion, existing algorithm have：1.Ryu et al. [2] is first with traditional quality evaluation method and synthesis tolerance method point
It is other that quality evaluation is carried out to texture image and depth image, respectively obtain two mass fractions.Finally two fractions are added
Power obtains the mass fraction of view picture composograph；2.Wang et al. [3] calculates composograph and original undistorted image first
Texture maps similarity image and depth map similarity image；Then normalizing is carried out to depth similarity image using texture information figure
Change；Finally, the depth similarity image after normalization and texture paging image are combined to the quality point to produce composograph
Number.For the second class distortion, existing method is：Bosc et al. [4] constructs visual angle composograph storehouse first, and
A kind of modified quality evaluating method is proposed on the basis of SSIM algorithms.The image library includes 7 kinds of DIBR algorithms and carries out visual angle
The image of synthesis；Only corresponding fringe region in the texture image and composograph at original visual angle is utilized in quality evaluation
SSIM is evaluated, finally using SSIM averages as final mass fraction；Conze et al. [5] is calculated using SSIM algorithms close first
Into the distortion map between image and original texture image, then three are calculated according to Texture complication, gradient direction and contrast
Weighted graph, is finally weighted distortion map processing using weighted graph, so as to obtain mass fraction.Open gorgeous grade [6] and be directed to visual angle
The characteristics of distortion at edge is frequently more obvious in composograph, by analyzing the pixel difference of composograph and original image,
And higher weights are assigned to edge pixel, and then obtain final mass fraction；Stankovic etc. [7] proposes small using morphology
Ripple carries out multi-level decomposition to original image and composograph, and calculates mean square error in multiple detail subbands, on this basis
Further calculate multiple dimensioned Y-PSNR and as mass fraction；The algorithm that Battisti etc. [8] is proposed is first to ginseng
Examine image and composograph carries out piecemeal, matched with motion estimation algorithm；Small echo change is carried out to the image block after matching
Simultaneously design factor histogram is changed, utilizes the distortion level of Kolmogorov-Smirnov distance description composographs；Jung etc. [9]
First main distortion zone is detected with the left and right multi-view image after synthesis and disparity map；Then to the distortion zone meter at two visual angles
SSIM fractions are calculated, finally the SSIM fractions at left and right visual angle be averaged as final mass fraction.This method lays particular emphasis on conjunction
Influence of the left and right visual angle asymmetry to synthesis quality during.Li et al. people [10] proposes one kind and is based on local geometric distortion
With the visual angle composograph quality evaluation of global clarity.First, detect hole region, then by the size of hole region and
Intensity combines and calculates local geometric distortion fraction；Then, the global clarity of image is calculated using method fuzzy again
Fraction.Finally, two fractions are combined to the mass fraction of generation composograph.Gu et al. [11] proposes one kind and is based on returning certainly
Return the composograph quality evaluating method of model.This method calculates the autoregression image of composograph first, then utilizes conjunction
Geometric distortion region is extracted into the difference of image and autoregression image.Using a threshold value, by the error image of two images
It is transformed to bianry image.Finally, the similarity figure between bianry image and the natural image predicted is by composite diagram the most
The mass fraction of picture.
Existing composograph quality evaluating method above has following defect：First, each method is both for it
In a type of distortion be designed, and have ignored another distortion brought in multi-view image building-up process.Therefore,
They can not the whole visual angle building-up process of effectively evaluating；In addition, in all of the above processes, only the method for Gu is no ginseng
Examine image quality evaluating method, i.e., this method without original image as reference.Other methods are full reference method, i.e., he
Have to rely on original image could to composograph carry out quality evaluation.And in reality, original image can not be often obtained, this
Constrain the application of existing full reference method.To sum up, there is an urgent need for design one kind to be evaluated for whole building-up process
Reference-free quality evaluation method.
[1]Y.C.Fan,P.K.Huang,and D.W.Shen,“3DTV depth map reconstruction
based on structured light scheme,”IEEE Int.Instrum.Meas.Technol.Conf.,pp.835-
838,May 2013.
[2]S.Ryu,S.Kim,and K.Sohn,“Synthesis quality prediction model based
on distortion intolerance,”IEEE Int.Conf.Image Process.,pp.585-589,Oct.2014.
[3]J.H.Wang,S.Q.Wang,K.Zeng and Z.Wang,“Quality assessment of multi-
view-plus-depth images,”IEEE International Conference on Multimedia and Expo,
pp.85-90,Jul.2017.
[4]E.Bosc,R.Pépion,P.L.Callet,M.Koppel,P.N.Nya,L.Morin and
M.Pressigout,“Towards a new qualtiy metric for 3-D synthesized view
assessment,”IEEE J.Select.Top.Signal Process.,vol.5,no.7,pp.1332-1343,
Sep.2011.
[5]P.H.Conze,P.Robert and L.Morin,“Objective view synthesis quality
assessment,”Electron.Imag.Int.Society for Optics and Photonics,vol.8288,
pp.8288-8256,Feb.2012.
[6] Zhang Yan, Anping, You Zhixiang, Zhang Zhaoyang, the virtual view image quality evaluation method based on Edge difference,《Electronics
With information journal》, 35 (8):1894-1900,2013.
[7]D.S.Stankovic,D.Kukolj and P.L.Callet,“DIBR synthesized image
quality assessment based on morphological wavelets,”IEEE Int.Workshop on
Quality of Multimedia Experience,pp.1-6,Jan.2015.
[8]F.Battisti,E.Bosc,M.Carli and P.L.Callet,“Objective image quality
assessment of 3D synthesized views,”Sig.Process.:Image Commun.,vol.30,pp.78-
88,Jan.2015.
[9]Y.J.Jung,H.G.Kim,and Y.M.Ro,“Critical binocular asymmetry measure
for perceptual quality assessment of synthesized stereo 3D images in view
synthesis”,IEEE Transactions on Circuits and Systems for Video Technology,26
(7):1201-1214,2016.
[10]L.D.Li,Y.Zhou,K.Gu,W.S.Lin,and S.Q.Wang,“Quality assessment of
DIBR-synthesized images by measuring local geometric distortions and global
sharpness,”IEEE Trans.Multimedia.
[11]K.Gu,V.Jakhetiya,J.F.Qiao,X.L.Li,W.S.Lin and D.Thalmann,“Model-
based referenceless quality metric of 3D synthesized images using local image
description,”IEEE Trans.Image Process.,vol.PP,pp.1-1,Jul.2017.DOI:10.1109/
TIP.2017.2733164.
The content of the invention
Goal of the invention：The process of visual angle synthesis mainly includes the acquisition of texture image and depth image, processing, transmission, void
Intend the drawing process at visual angle.Distortion can be introduced during these, wherein, transmission is got from texture and depth image,
Traditional distortion can be introduced, such as fuzzy, blocking effect etc.；It is imperfect due to rendering algorithm in the drawing process of virtual perspective,
Drafting distortion can be brought in the visual angle newly synthesized.To sum up, two kinds of distortions are co-existed in the visual angle finally synthesized, i.e., traditional distortion
With drafting distortion.And existing method is designed for one of which distortion, it is impossible to which effectively evaluating entirely regards
Two class distortions in the building-up process of angle.In order to solve the above technical problems, the present invention proposes a kind of virtual perspective synthesis view image quality
Amount without with reference to evaluation method, for the present invention by analyzing the characteristics of two class distortions, design feature quantifies the mistake in composograph
Very：The distortion brought first to the edge damage of image and the unnatural property of texture quantify, and extract individual features, then
All features are integrated, the evaluation model for the distortion that can be brought to whole building-up process is found using machine learning.
Technical solution：To realize above-mentioned technique effect, technical solution proposed by the present invention is：
A kind of non-reference picture quality appraisement method towards visual angle synthesis, including step：
(1) one group of visual angle composograph is collected, forms visual angle composograph storehouse；
(2) the step of distortion in quantization visual angle composograph storehouse in the composograph of each width visual angle, quantization, is included to every
Width visual angle composograph performs step (2-1) to (2-3)：
(2-1) definition visual angle composograph is scalogram as 1；Carry out n-1 Gauss low pass respectively to visual angle composograph
Filtering, and remember that the filter result of ith is scalogram as i+1, i ∈ [1,2 ..., n-1]；Scalogram is formed with n as 1 to n
The metric space of scale；
(2-2) builds DoG models；DoG models include n image, are denoted as DoG1,DoG2,…,DoGn；Wherein, DoGiFor
Scalogram is as i+1 and scalogram are as the difference of i, i ∈ [1,2 ..., n-1]；DoGnIt is scalogram as n；
(12-3) carries out characteristic parameter extraction respectively to each image in DoG models, obtains 7 edges of each image
Set direction characteristic parameter and 2 texture naturality characteristic parameters；
(3) image in the composograph storehouse of visual angle is randomly divided into training image and test image two parts；Using random
Forest law is modeled the characteristic parameter of training image, obtains Environmental Evaluation Model；Using the characteristic parameter of test image as matter
The input of evaluation model is measured, obtains the objective quality scores of test image.
Further, any one width scalogram is expressed as i：
Wherein, * represents convolution symbol, σiRepresent scalogram as the standard deviation of the gaussian kernel function of i；Li(x, y) represents scale
The pixel value at pixel (x, y) place in image i；I (x, y) represents the pixel value at pixel (x, y) place in the composograph of visual angle；G
() is gaussian kernel function, G (x, y, σi) expression formula be：
Further, any one width graphical representation is in the DoG models：
In formula, DoGi(x, y) represents image DoGiThe pixel value at middle pixel (x, y) place.
Further, any one image DoG in the model to DoGiCarry out edge direction selectional feature parameter
The step of extraction, includes：
1) DoG that changes commanders is become using overcomplete waveletiDecompose on 2 scales and 6 directions：
2) wavelet coefficient of same direction different scale is included into a set, 6 wavelet coefficient set is obtained, are denoted as
Zit, t=[1,2 ..., 6]；
3) step S3-1 to S3-4 is performed to each wavelet coefficient set：
S3-1：Calculate ZitSingle order absolute moment：
In formula, J1For ZitSingle order absolute moment, z is stochastic variable, and θ represents gamma function,
γitRepresent ZitForm parameter, y2For an intermediate parameters,σ represents the standard deviation of wavelet coefficient；
Order-| zy2|γit=Yit, obtainWillSubstitute into J1Meter
Formula is calculated, is obtained：
S3-2：Calculate ZitSecond moment be：
J2=σ2 (6)
S3-3：Orderγ is calculated according to formula (7)it：
In formula, zjRepresent set ZitIn j-th of wavelet coefficient, h represent ZitThe number of middle wavelet coefficient；
4) by 6 wavelet coefficient set Zi1To Zi6A set is merged into, is denoted as Zi7；To Zi7Step 3) is performed, is obtained
γi7；γi1,γi2,…,γi7As DoGi7 edge direction selectional feature parameters.
Further, to any one DoG image DoGiThe step of carrying out texture naturality characteristic parameter extraction is wrapped
Include：
(5-1) calculates DoG images DoGiGradient image gi,giMeet：
In formula, gi(x, y) represents gradient image giThe pixel value at middle pixel (x, y) place,WithNot
Represent DoGiHorizontal direction gradient and vertical gradient；Wherein,
(5-2) is to DoG images DoGiAnd its gradient image giIt is normalized：
Wherein,Represent gradient image giThe pixel value at pixel (p, q) place after normalization；Represent DoG
Image DoGiThe pixel value at pixel (p, q) place after normalization；INT represents floor operation；gi,minRepresent gradient image giMost
Small pixel value, gi,maxRepresent gradient image giMax pixel value, Ni,gRepresent gradient image giMaximum gray scale after normalization
Value；DoGi,minRepresent DoG images DoGiMinimum pixel value, DoGi,maxRepresent DoG images DoGiMax pixel value, Ni,DTable
Show DoG images DoGiMaximum gradation value after normalization；
(5-3) calculates DoG images DoG according to the result of step (5-2)iGray level-gradient co-occurrence matrix Mi, MiIn element
It is expressed as Mi(p, q), MiThe value of (p, q) is：MeetAndPixel number；
(5-4) is from Gray level-gradient co-occurrence matrix MiMiddle extraction energy and gradient mean square deviation are as image DoGi2 textures from
Right property characteristic parameter.
Further, from Gray level-gradient co-occurrence matrix M in the step (5-4)iIt is middle extraction energy computational methods be：
From Gray level-gradient co-occurrence matrix MiIt is middle extraction gradient mean square deviation computational methods be：
Wherein,Represent gradient image giAverage.
Beneficial effect：Compared with prior art, the present invention has the advantage that：
1st, the shortcomings that existing method can only evaluate a kind of distortion in building-up process is overcome, this method can be commented effectively
Two class distortions in the whole building-up process of valency；
2nd, performance of the invention is substantially better than existing without visual quality evaluation method is referred to, and is specifically included：General nothing
Reference image quality appraisement method, existing visual angle composograph quality evaluating method；
3 compare with existing method, and inter-library performance of the invention is best, and scalability is most strong.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention.
Embodiment
The present invention is further described below in conjunction with the accompanying drawings.
Fig. 1 show the flow chart of the present invention, as seen from the figure：The overall flow of the present invention is divided into four module：
1st, scale-space representation 2, DoG model foundations 3, feature extraction 4, training quality evaluation model.Below to this four
Step describes in detail：
Module 1：Scale-space representation-and for a width visual angle composograph, image is carried out using gauss low frequency filter
Repeatedly filtering, so as to construct the metric space of image.
Module 2：The foundation of the DOG models-image work of adjacent scale is poor, obtains the error image of gaussian filtering, i.e.,
DoG images.
Module 3：Feature extraction-carry out feature extraction in DoG images and last scalogram picture, in each width figure
The feature specifically extracted as in is 7 edge direction selectional features and 2 texture naturality features.
Module 4：The foundation of Environmental Evaluation Model-by the image in database is randomly divided into two parts.Part I is used for
Training quality evaluation model, Part II are used for testing.All training characteristics in Part I are inputted, are trained using random forest
Mass evaluation model.Then prediction of quality is carried out to the test image in Part II using the model of training, is used for
The objective quality scores of assessment design composograph quality.
Technical scheme is explained in detail with reference to specific embodiment：
Step 1：Due to the two class distortions brought in building-up process can cause edge destruction and texture it is unnatural
Property, and DoG decomposition can effectively capture edge and textural characteristics in image.Needed since DoG is decomposed in adjacent scalogram picture
Upper progress, so, the foundation of metric space is carried out first by Gaussian filter：
Definition visual angle composograph is scalogram as 1；N-1 Gassian low-pass filter is carried out respectively to visual angle composograph,
And remember that the filter result of ith is scalogram as i+1, i ∈ [1,2 ..., n-1]；Scalogram has n scale as 1 to n formation
Metric space；Any one width scalogram is expressed as i：
Wherein, * represents convolution symbol, σiRepresent scalogram as the standard deviation of the gaussian kernel function of i；Li(x, y) represents scale
The pixel value at pixel (x, y) place in image i；I (x, y) represents the pixel value at pixel (x, y) place in the composograph of visual angle；G
() is gaussian kernel function, G (x, y, σi) expression formula be：
In the present embodiment, the value of n is 4, so constructing the metric space containing four scalogram pictures, is represented respectively
For：L1,L2,L3,L4。
Step 2：Build DoG models；DoG models include 4 DoG images, are denoted as DoG1,DoG2,DoG3,DoG4；Wherein,
DoGiIt is scalogram as i+1 and scalogram are as the difference of i, i ∈ [1,2 ..., 3]；DoG4It is scalogram as the expression of 4, DoG images
Formula is：
DoGi+1(x, y)=Li(x,y)-Li+1(x,y),i∈[1,2,3]
DoG4(x, y)=L4(x,y)
In formula, DoGi+1(x, y) represents pixel values of the DoG images i+1 at pixel (x, y) place.
Step 3：For a width natural image, the wavelet coefficient on different scale equidirectional meets Generalized Gaussian point
Cloth, this rule are called set direction statistics.And the distortion in composograph can cause the statistics of this natural image special
Property.Become each width DoG picture breakdowns of changing commanders using overcomplete wavelet on 2 scales and 6 directions, then to DoG images
Wavelet coefficient carries out discrete normalization, and uses Generalized Gaussian to the wavelet coefficient on the equidirectional different scale after normalization
Distribution function is fitted.
To any one image DoG in DoG modelsiThe step of carrying out edge direction selectional feature parameter extraction is wrapped
Include：
S1 becomes the DoG that changes commanders using overcomplete waveletiDecompose on 2 scales and 6 directions：
The wavelet coefficient of same direction different scale is included into a set by S2, since each width DoG images are divided into 6
Direction, 2 scales, therefore 6 wavelet coefficient set are obtained, it is denoted as Zit, t=[1,2 ..., 6]；
S3 performs step S3-1 to S3-4 to each wavelet coefficient set：
S3-1：Calculate ZitSingle order absolute moment：
In formula, J1For ZitSingle order absolute moment, z is stochastic variable, and θ represents gamma function,
γitRepresent ZitForm parameter, y2For an intermediate parameters,
Order-| zy2|γit=Yit, obtainWillSubstitute into J1Calculating it is public
Formula, obtains：
S3-2：Calculate ZitSecond moment be：
J2=σ2 (6)
S3-3：Orderγ is calculated according to formula (7)it：
In formula, zjRepresent set ZitIn j-th of wavelet coefficient, h represent ZitThe number of middle wavelet coefficient；
S4 is by 6 wavelet coefficient set Zi1To Zi6A set is merged into, is denoted as Zi7；To Zi7Step S3 is performed, is obtained
γi7；γi1,γi2,…,γi7As DoGi7 edge direction selectional feature parameters.
In the present embodiment, 4 width DoG images, which amount to, obtains 28 edge direction selectional feature parameters.
Step 4：Due to equally bringing the loss of texture naturality in the building-up process of visual angle, so we use gray scale
Gradient co-occurrence matrix describes the textural characteristics of image.
To any one DoG image DoGiThe step of carrying out texture naturality characteristic parameter extraction includes：
(5-1) calculates DoG images DoGiGradient image gi,giMeet：
In formula, gi(x, y) represents gradient image giThe pixel value at middle pixel (x, y) place,WithNot
Represent DoGiHorizontal direction gradient and vertical gradient；Wherein,
(5-2) is to image DoGiAnd its gradient image giIt is normalized：
Wherein,Represent gradient image giThe pixel value at pixel (p, q) place after normalization；Represent DoG
Image DoGiThe pixel value at pixel (p, q) place after normalization；INT represents floor operation；gi,minRepresent gradient image giMost
Small pixel value, gi,maxRepresent gradient image giMax pixel value, Ni,gRepresent gradient image giMaximum gray scale after normalization
Value；DoGi,minRepresent DoG images DoGiMinimum pixel value, DoGi,maxRepresent DoG images DoGiMax pixel value, Ni,DTable
Show DoG images DoGiMaximum gradation value after normalization；
(5-3) calculates DoG images DoG according to the result of step (5-2)iGray level-gradient co-occurrence matrix Mi, MiIn element
It is expressed as Mi(p, q), MiThe value of (p, q) is：MeetAndPixel number；
(5-4) is from Gray level-gradient co-occurrence matrix MiMiddle extraction energy and gradient mean square deviation are as image DoGi2 textures from
Right property characteristic parameter；From Gray level-gradient co-occurrence matrix MiIt is middle extraction energy computational methods be：
From Gray level-gradient co-occurrence matrix MiIt is middle extraction gradient mean square deviation computational methods be：
Wherein,Represent gradient image giAverage.
Each width DoG images can extract 2 and represent the feature of texture naturality, so 2 × 4=8 texture is obtained
Naturality characteristic parameter.
Step 5：For the visual angle composograph of any one width input, 28 set direction features and 8 can be obtained
Texture naturality feature, shares 36 characteristic parameters.Image in the composograph storehouse of visual angle is randomly divided into training image by us
With test image two parts；The characteristic parameter of training image is modeled using random forest method, obtains Environmental Evaluation Model；
Input using the characteristic parameter of test image as Environmental Evaluation Model, obtains the objective quality scores of test image.
The technique effect of the present invention is illustrated below by specific test data.
The experimental section of the present invention is carrying out disclosed in 2 on the composograph data set of visual angle.That is MCL storehouses and IVC-
DIBR storehouses.684 pairs of visual angle composographs, i.e. 684 LOOK LEFT images and 684 LOOK RIGHT images are included in MCL storehouses.The data
Image in storehouse, has 648 pairs comprising traditional distortion, distortion is drawn at the visual angle that remaining 36 pairs of images include.IVC is provided in storehouse
12 width original images and 84 width comprise only the image that distortion is drawn at visual angle.It is to be tested for one in the operating process of experiment
Visual angle composograph database, image is divided into 80% and 20% at random first, wherein, 80% image is used for carrying out model
Foundation, 20% image is used for the test of model.In order to avoid the generation of contingency, this process circulates 1000 times, 4 individual characteies
The intermediate value of energy index is by as final performance parameter.
First, we by the present invention performance and existing visual angle composograph quality evaluating method in two databases
It is compared.PLCC/SRCC/KRCC numerical value is bigger, and RMSE numerical value is smaller, illustrates that algorithm performance is better.In table 1：Related coefficient
(Pearson linear correlation coefficient, PLCC) is the linearly dependent coefficient after nonlinear regression；
Square error root (Root mean squared error, RMSE) is the standard deviation after nonlinear regression；Kendall grades are related
Coefficient (Kendall ' s Rank Correlation Coefficient, KRCC)；Spearman related coefficients (Spearman
Rank order correlation coefficient, SRCC).Since the method for Wang relies on depth image, and IVC storehouses do not have
Depth image is provided with, so performance of this method on IVC storehouses can not be evaluated.Therefore, in table 1, with strigula "-" mark
Note.
The performance comparison of 1 method provided by the present invention of table and existing visual angle composograph quality evaluation algorithm
As shown in Table 1, PLCC/SRCC/KRCC of the method provided by the present invention on two storehouses apparently higher than it is all its
His algorithm, RMSE are minimum.This illustrates that method performance provided by the present invention has obvious superiority.And due to MCL
Image in storehouse contains two kinds of distortion, this further proves that method provided by the present invention can synthesize whole visual angle
During distortion carry out the most accurately evaluation.
In order to further verify the performance of method provided by the present invention, we are by method provided by the present invention and general
Image quality evaluating method is compared.General image quality evaluation algorithm refers to the type of distortion that need not know image, you can
The algorithm of quality evaluation is carried out to image.
The performance comparison sheet of 2 method provided by the present invention of table and general image quality evaluation algorithm
From the data of table 2, the performance of method provided by the present invention is substantially better than general image quality evaluation algorithm.
It is embodied in best forecasting accuracy and monotonicity.
For based on trained quality evaluation algorithm, inter-library performance, you can autgmentability is that the important evaluation of these methods refers to
Mark.Inter-library experiment, refers to train Environmental Evaluation Model with the feature of all image zooming-outs in an image library, then utilizes
The model tests the performance of all images in another database.Based on this, to all based on trained quality evaluation side
Method carries out the confirmatory experiment of scalability.
Inter-library performance comparison table of the table 3 based on trained algorithm
From the experimental result of table 3, method provided by the present invention has best inter-library performance, i.e., the present invention is carried
The scalability of the method for confession is most strong.
To sum up, all experimental results prove the superiority of method provided by the present invention.
The above is only the preferred embodiment of the present invention, it should be pointed out that：For the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (6)
1. a kind of non-reference picture quality appraisement method towards visual angle synthesis, it is characterised in that including step：
(1) one group of visual angle composograph is collected, forms visual angle composograph storehouse；
(2) the step of distortion in quantization visual angle composograph storehouse in the composograph of each width visual angle, quantization, includes regarding every width
Angle composograph performs step (2-1) to (2-3)：
(2-1) definition visual angle composograph is scalogram as 1；N-1 Gassian low-pass filter is carried out respectively to visual angle composograph,
And remember that the filter result of ith is scalogram as i+1, i ∈ [1,2 ..., n-1]；Scalogram has n scale as 1 to n formation
Metric space；
(2-2) builds DoG models；DoG models include n image, are denoted as DoG1,DoG2,…,DoGn；Wherein, DoGiFor scalogram
As i+1 and scalogram are as the difference of i, i ∈ [1,2 ..., n-1]；DoGnIt is scalogram as n；
(12-3) carries out characteristic parameter extraction respectively to each image in DoG models, obtains 7 edge directions of each image
Selectional feature parameter and 2 texture naturality characteristic parameters；
(3) image in the composograph storehouse of visual angle is randomly divided into training image and test image two parts；Using random forest
Method is modeled the characteristic parameter of training image, obtains Environmental Evaluation Model；Commented using the characteristic parameter of test image as quality
The input of valency model, obtains the objective quality scores of test image.
A kind of 2. non-reference picture quality appraisement method towards visual angle synthesis according to claim 1, it is characterised in that
Any one width scalogram is expressed as i：
<mrow>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>*</mo>
<mi>G</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>,</mo>
<msub>
<mi>&sigma;</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>i</mi>
<mo>&Element;</mo>
<mo>&lsqb;</mo>
<mn>2</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>n</mi>
<mo>&rsqb;</mo>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, * represents convolution symbol, σiRepresent scalogram as the standard deviation of the gaussian kernel function of i；Li(x, y) represents scalogram picture
The pixel value at pixel (x, y) place in i；I (x, y) represents the pixel value at pixel (x, y) place in the composograph of visual angle；G () is
Gaussian kernel function, G (x, y, σi) expression formula be：
<mrow>
<mi>G</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>,</mo>
<msub>
<mi>&sigma;</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mn>2</mn>
<msubsup>
<mi>&pi;&sigma;</mi>
<mi>i</mi>
<mn>2</mn>
</msubsup>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msup>
<mi>x</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>y</mi>
<mn>2</mn>
</msup>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mn>2</mn>
<msup>
<msub>
<mi>&sigma;</mi>
<mi>i</mi>
</msub>
<mn>2</mn>
</msup>
</mrow>
</msup>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
A kind of 3. non-reference picture quality appraisement method towards visual angle synthesis according to claim 2, it is characterised in that
Any one width graphical representation is in the DoG models：
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>DoG</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>L</mi>
<mrow>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>i</mi>
<mo>&Element;</mo>
<mo>&lsqb;</mo>
<mn>1</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>n</mi>
<mo>-</mo>
<mn>1</mn>
<mo>&rsqb;</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>DoG</mi>
<mi>n</mi>
</msub>
<mrow>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>L</mi>
<mi>n</mi>
</msub>
<mrow>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula, DoGi(x, y) represents image DoGiThe pixel value at middle pixel (x, y) place.
A kind of 4. non-reference picture quality appraisement method towards visual angle synthesis according to claim 3, it is characterised in that
Any one image DoG in the model to DoGiThe step of carrying out edge direction selectional feature parameter extraction includes：
1) DoG that changes commanders is become using overcomplete waveletiDecompose on 2 scales and 6 directions：
2) wavelet coefficient of same direction different scale is included into a set, 6 wavelet coefficient set is obtained, are denoted as Zit, t
=[1,2 ..., 6]；
3) step S3-1 to S3-4 is performed to each wavelet coefficient set：
S3-1：Calculate ZitSingle order absolute moment：
<mrow>
<msub>
<mi>J</mi>
<mn>1</mn>
</msub>
<mo>=</mo>
<msubsup>
<mo>&Integral;</mo>
<mrow>
<mo>-</mo>
<mi>&infin;</mi>
</mrow>
<mrow>
<mo>+</mo>
<mi>&infin;</mi>
</mrow>
</msubsup>
<mo>|</mo>
<mi>z</mi>
<mo>|</mo>
<mfrac>
<mrow>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<msub>
<mi>y</mi>
<mn>2</mn>
</msub>
</mrow>
<mrow>
<mn>2</mn>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mo>|</mo>
<msub>
<mi>zy</mi>
<mn>2</mn>
</msub>
<msup>
<mo>|</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
</msup>
</mrow>
</msup>
<mi>d</mi>
<mi>z</mi>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<msub>
<mi>y</mi>
<mn>2</mn>
</msub>
</mrow>
<mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<msubsup>
<mo>&Integral;</mo>
<mn>0</mn>
<mrow>
<mo>+</mo>
<mi>&infin;</mi>
</mrow>
</msubsup>
<mo>|</mo>
<mi>z</mi>
<mo>|</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mo>|</mo>
<msub>
<mi>zy</mi>
<mn>2</mn>
</msub>
<msup>
<mo>|</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
</msup>
</mrow>
</msup>
<mi>d</mi>
<mi>z</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula, J1For ZitSingle order absolute moment, z is stochastic variable, and θ represents gamma function,
γitRepresent ZitForm parameter, y2For an intermediate parameters,σ represents the standard deviation of wavelet coefficient；
Order-| zy2|γit=Yit, obtainWillSubstitute into J1Calculating it is public
Formula, obtains：
<mrow>
<msub>
<mi>J</mi>
<mn>1</mn>
</msub>
<mo>=</mo>
<mi>&sigma;</mi>
<mfrac>
<mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<msqrt>
<mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
S3-2：Calculate ZitSecond moment be：
J2=σ2 (6)
S3-3：Orderγ is calculated according to formula (7)it：
<mrow>
<mfrac>
<msubsup>
<mi>J</mi>
<mn>1</mn>
<mn>2</mn>
</msubsup>
<msub>
<mi>J</mi>
<mn>2</mn>
</msub>
</mfrac>
<mo>=</mo>
<mfrac>
<mrow>
<msup>
<mi>&theta;</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mn>2</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
</mrow>
<mo>)</mo>
</mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mrow>
<mn>3</mn>
<mo>/</mo>
<msub>
<mi>&gamma;</mi>
<mrow>
<mi>i</mi>
<mi>t</mi>
</mrow>
</msub>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>=</mo>
<mfrac>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mfrac>
<mn>1</mn>
<mi>h</mi>
</mfrac>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>h</mi>
</msubsup>
<mo>|</mo>
<msub>
<mi>z</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mrow>
<mfrac>
<mn>1</mn>
<mi>h</mi>
</mfrac>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>h</mi>
</msubsup>
<mo>|</mo>
<msub>
<mi>z</mi>
<mi>j</mi>
</msub>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>7</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula, zjRepresent set ZitIn j-th of wavelet coefficient, h represent ZitThe number of middle wavelet coefficient；
4) by 6 wavelet coefficient set Zi1To Zi6A set is merged into, is denoted as Zi7；To Zi7Step 3) is performed, obtains γi7；
γi1,γi2,…,γi7As DoGi7 edge direction selectional feature parameters.
A kind of 5. non-reference picture quality appraisement method towards visual angle synthesis according to claim 4, it is characterised in that
To any one DoG image DoGiThe step of carrying out texture naturality characteristic parameter extraction includes：
(5-1) calculates DoG images DoGiGradient image gi,giMeet：
<mrow>
<msub>
<mi>g</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msqrt>
<mrow>
<mo>&dtri;</mo>
<msub>
<mi>h</mi>
<mi>i</mi>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mo>&dtri;</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<mo>,</mo>
<mi>i</mi>
<mo>=</mo>
<mo>&lsqb;</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>n</mi>
<mo>&rsqb;</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>8</mn>
<mo>)</mo>
</mrow>
</mrow>
In formula, gi(x, y) represents gradient image giThe pixel value at middle pixel (x, y) place,WithDo not represent
DoGiHorizontal direction gradient and vertical gradient；Wherein,
<mrow>
<msub>
<mi>h</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>Do</mi>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>+</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>Do</mi>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mi></mi>
<mrow>
<mo>(</mo>
<mn>9</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>Do</mi>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>Do</mi>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mi></mi>
<mrow>
<mo>(</mo>
<mn>10</mn>
<mo>)</mo>
</mrow>
</mrow>
(5-2) is to DoG images DoGiAnd its gradient image giIt is normalized：
<mrow>
<msub>
<mover>
<mi>g</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>I</mi>
<mi>N</mi>
<mi>T</mi>
<mo>&lsqb;</mo>
<mfrac>
<mrow>
<msub>
<mi>g</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>g</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>g</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>g</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>min</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>&times;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>N</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>g</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>11</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mover>
<mi>D</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>I</mi>
<mi>N</mi>
<mi>T</mi>
<mo>&lsqb;</mo>
<mfrac>
<mrow>
<msub>
<mi>DoG</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>DoG</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>DoG</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>DoG</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>N</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>D</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>12</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,Represent gradient image giThe pixel value at pixel (p, q) place after normalization；Represent DoG images
DoGiThe pixel value at pixel (p, q) place after normalization；INT represents floor operation；gi,minRepresent gradient image giMinimum image
Element value, gi,maxRepresent gradient image giMax pixel value, Ni,gRepresent gradient image giMaximum gradation value after normalization；
DoGi,minRepresent DoG images DoGiMinimum pixel value, DoGi,maxRepresent DoG images DoGiMax pixel value, Ni,DRepresent
DoG images DoGiMaximum gradation value after normalization；
(5-3) calculates DoG images DoG according to the result of step (5-2)iGray level-gradient co-occurrence matrix Mi, MiIn element representation
For Mi(p, q), MiThe value of (p, q) is：MeetAndPixel number；
(5-4) is from Gray level-gradient co-occurrence matrix MiMiddle extraction energy and gradient mean square deviation are as image DoGi2 texture naturalities
Characteristic parameter.
A kind of 6. non-reference picture quality appraisement method towards visual angle synthesis according to claim 5, it is characterised in that
From Gray level-gradient co-occurrence matrix M in the step (5-4)iIt is middle extraction energy computational methods be：
<mrow>
<msub>
<mi>E</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>p</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<msub>
<mi>N</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>D</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>q</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<msub>
<mi>N</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>g</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<msup>
<mrow>
<mo>&lsqb;</mo>
<msub>
<mi>M</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>13</mn>
<mo>)</mo>
</mrow>
</mrow>
From Gray level-gradient co-occurrence matrix MiIt is middle extraction gradient mean square deviation computational methods be：
<mrow>
<msub>
<mi>G</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>q</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<msub>
<mi>N</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>g</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<mi>q</mi>
<mo>-</mo>
<msub>
<mover>
<mi>g</mi>
<mo>&OverBar;</mo>
</mover>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>p</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<msub>
<mi>N</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>D</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<msup>
<mrow>
<mo>&lsqb;</mo>
<msub>
<mi>M</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
<mrow>
<mn>1</mn>
<mo>/</mo>
<mn>2</mn>
</mrow>
</msup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>14</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,Represent gradient image giAverage.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711399720.XA CN107977967B (en) | 2017-12-22 | 2017-12-22 | No-reference image quality evaluation method for view angle synthesis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711399720.XA CN107977967B (en) | 2017-12-22 | 2017-12-22 | No-reference image quality evaluation method for view angle synthesis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107977967A true CN107977967A (en) | 2018-05-01 |
| CN107977967B CN107977967B (en) | 2022-05-03 |
Family
ID=62007424
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711399720.XA Active CN107977967B (en) | 2017-12-22 | 2017-12-22 | No-reference image quality evaluation method for view angle synthesis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107977967B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712134A (en) * | 2018-12-28 | 2019-05-03 | 武汉虹识技术有限公司 | Iris image quality evaluation method, device and electronic equipment |
| CN110211090A (en) * | 2019-04-24 | 2019-09-06 | 西安电子科技大学 | A method of for assessment design composograph quality |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102611910A (en) * | 2011-01-19 | 2012-07-25 | 北京东方文骏软件科技有限责任公司 | Objective evaluation method of no-reference video quality weighted based by key frame image quality |
| US20120257164A1 (en) * | 2011-04-07 | 2012-10-11 | The Chinese University Of Hong Kong | Method and device for retinal image analysis |
| CN103581661A (en) * | 2013-10-28 | 2014-02-12 | 宁波大学 | Method for evaluating visual comfort degree of three-dimensional image |
| CN105744256A (en) * | 2016-03-31 | 2016-07-06 | 天津大学 | Three-dimensional image quality objective evaluation method based on graph-based visual saliency |
| CN105898278A (en) * | 2016-05-26 | 2016-08-24 | 杭州电子科技大学 | Stereoscopic video saliency detection method based on binocular multidimensional perception characteristic |
| CN107343196A (en) * | 2017-07-18 | 2017-11-10 | 天津大学 | One kind mixing distortion non-reference picture quality appraisement method |
-
2017
- 2017-12-22 CN CN201711399720.XA patent/CN107977967B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102611910A (en) * | 2011-01-19 | 2012-07-25 | 北京东方文骏软件科技有限责任公司 | Objective evaluation method of no-reference video quality weighted based by key frame image quality |
| US20120257164A1 (en) * | 2011-04-07 | 2012-10-11 | The Chinese University Of Hong Kong | Method and device for retinal image analysis |
| CN103581661A (en) * | 2013-10-28 | 2014-02-12 | 宁波大学 | Method for evaluating visual comfort degree of three-dimensional image |
| CN105744256A (en) * | 2016-03-31 | 2016-07-06 | 天津大学 | Three-dimensional image quality objective evaluation method based on graph-based visual saliency |
| CN105898278A (en) * | 2016-05-26 | 2016-08-24 | 杭州电子科技大学 | Stereoscopic video saliency detection method based on binocular multidimensional perception characteristic |
| CN107343196A (en) * | 2017-07-18 | 2017-11-10 | 天津大学 | One kind mixing distortion non-reference picture quality appraisement method |
Non-Patent Citations (6)
| Title |
|---|
| LI, L.,ET.AL: "No-reference quality assessment of deblocked images", 《NEUROCOMPUTING》 * |
| PEI, S. C.,ET.AL: "Image quality assessment using human visual DOG model fused with random forest", 《TRANSACTIONS ON IMAGE PROCESSING》 * |
| WANG Z, ET.AL: "Image quality assessment: from error visibility to structural similarity", 《 IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 卢兆林: "基于偏微分方程理论的图像复原技术研究", 《中国优秀硕士学位论文全文数据库》 * |
| 吴东: "基于感知特征的图像质量评价方法研究", 《中国优秀硕士学位论文全文数据库》 * |
| 贾永红: "《数字图像处理》", 31 July 2015, 武汉大学出版社 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712134A (en) * | 2018-12-28 | 2019-05-03 | 武汉虹识技术有限公司 | Iris image quality evaluation method, device and electronic equipment |
| CN109712134B (en) * | 2018-12-28 | 2020-11-06 | 武汉虹识技术有限公司 | Iris image quality evaluation method and device and electronic equipment |
| CN110211090A (en) * | 2019-04-24 | 2019-09-06 | 西安电子科技大学 | A method of for assessment design composograph quality |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107977967B (en) | 2022-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | No-reference and robust image sharpness evaluation based on multiscale spatial and spectral features | |
| Li et al. | No-reference image blur assessment based on discrete orthogonal moments | |
| Gao et al. | Image quality assessment based on multiscale geometric analysis | |
| CN106462771A (en) | 3D image significance detection method | |
| CN105338343B (en) | It is a kind of based on binocular perceive without refer to stereo image quality evaluation method | |
| CN104751456B (en) | Blind image quality evaluating method based on conditional histograms code book | |
| CN106228528B (en) | A kind of multi-focus image fusing method based on decision diagram and rarefaction representation | |
| CN107194872A (en) | Remote sensed image super-resolution reconstruction method based on perception of content deep learning network | |
| CN105049851B (en) | General non-reference picture quality appraisement method based on Color perception | |
| CN102629374B (en) | Image super resolution (SR) reconstruction method based on subspace projection and neighborhood embedding | |
| CN108053396A (en) | A kind of more distorted image quality without with reference to evaluation method | |
| CN106709958A (en) | Gray scale gradient and color histogram-based image quality evaluation method | |
| Yue et al. | Blind stereoscopic 3D image quality assessment via analysis of naturalness, structure, and binocular asymmetry | |
| CN109345502B (en) | Stereo image quality evaluation method based on disparity map stereo structure information extraction | |
| Zhou et al. | Blind quality index for multiply distorted images using biorder structure degradation and nonlocal statistics | |
| Yang et al. | Stereoscopic video quality assessment based on 3D convolutional neural networks | |
| CN104036493B (en) | No-reference image quality evaluation method based on multifractal spectrum | |
| Bhateja et al. | Fast SSIM index for color images employing reduced-reference evaluation | |
| CN108765414A (en) | Based on wavelet decomposition and natural scene count without referring to stereo image quality evaluation method | |
| CN106650572A (en) | Method for assessing quality of fingerprint image | |
| CN106485741A (en) | A kind of method of the non-firm point set registration retaining partial structurtes | |
| CN112950596A (en) | Tone mapping omnidirectional image quality evaluation method based on multi-region and multi-layer | |
| CN109344860A (en) | A kind of non-reference picture quality appraisement method based on LBP | |
| CN107977967A (en) | A kind of non-reference picture quality appraisement method towards visual angle synthesis | |
| CN104408694B (en) | Denoising method for nonlocal average image based on soft threshold |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Li Leida Inventor after: Zhou Yu Inventor after: Lu Zhaolin Inventor before: Zhou Yu Inventor before: Li Leida Inventor before: Lu Zhaolin |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |