CN102930545A - Statistical measure method for image quality blind estimation - Google Patents

Statistical measure method for image quality blind estimation Download PDF

Info

Publication number
CN102930545A
CN102930545A CN2012104406180A CN201210440618A CN102930545A CN 102930545 A CN102930545 A CN 102930545A CN 2012104406180 A CN2012104406180 A CN 2012104406180A CN 201210440618 A CN201210440618 A CN 201210440618A CN 102930545 A CN102930545 A CN 102930545A
Authority
CN
China
Prior art keywords
coefficient
image
dnt
estimation
wavelet
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.)
Pending
Application number
CN2012104406180A
Other languages
Chinese (zh)
Inventor
黄虹
张建秋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fudan University
Original Assignee
Fudan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Fudan University filed Critical Fudan University
Priority to CN2012104406180A priority Critical patent/CN102930545A/en
Publication of CN102930545A publication Critical patent/CN102930545A/en
Pending legal-status Critical Current

Links

Images

Abstract

The invention belongs to the technical field of image quality estimation and particularly relates to a statistical measure method for image quality blind estimation. According to the method, image distortion is described through K-L statistical measure between statistical distribution and fitting Gaussian distribution of divisive normalization transform (DNT) coefficients of distorted images in a wavelet domain. Analyses indicates that the K-L statistical measure for describing the image distortion does not need any reference or prior information from undistorted images, so that the problem that the existing image quality objective blind estimation method depends on feature selection and extraction, machine learning and the like can be solved. The estimation results in a LIVE image quality estimation database indicate that image quality estimation results through the blind estimation statistical measure are highly consistent with subjective estimation results in the database and superior to the blind estimation measure reported in documents.

Description

The statistical measurement method of the blind assessment of a kind of picture quality
Technical field
The invention belongs to the image quality measure technical field, be specifically related to the statistical measurement method of the blind assessment of picture quality.
Background technology
The obtaining of image, compress, in storage, transmission and the reproduction process, tend to digital picture is introduced the distortion of a large amount of dissimilar and different stages, these distortions will cause the degradation of picture quality.How picture quality being carried out accurately objective evaluation has become in the image processing field one and has had challenging problem [1] [2]
The objective evaluation research of picture quality, its purpose designs a kind of evaluation measure or algorithm exactly, and expectation obtains the result similar with the assessment of human eye subjective picture quality.Existing picture quality objective evaluation method mainly is full reference evaluation method, namely contrasts the quality of distorted image and reference picture and then obtains assessment result by a certain method [3] [4]Such appraisal procedure has limited its range of application, and for example: the image that digital image apparatus photographs does not generally have reference picture, and will expect that to the image of mobile phone or TV perhaps its reference picture is inconvenient by long Distance Transmission.This means in many application scenarios, exist the demand of double reference evaluation method (depending on a part of reference information) and blind appraisal procedure (without any need for reference information).The present invention provides a kind of new method for the blind assessment of picture quality.Because without any the reference information of relevant not distorted image, therefore the realization of blind assessment algorithm is faced with great difficulty in blind assessment [5]-[8]
In recent years, because natural scene statistics (Natural Scene Statistics, NSS) model can be described the statistical property that has nothing to do with content in the natural image, therefore be subject to more and more scholars' attention [3]-[5], [11], [12]The NSS model is the statistical model that natural scene is set up.Relevant studies show that: the edge of undistorted natural image wavelet conversion coefficient or joint probability distribution, can mix (Gaussian scale mixture by Gauss's yardstick, GSM) distributed model is described, and the wavelet coefficient of distorted image does not then possess this character [6] [10] [11]
If length is NRandom vector Can be expressed as
Figure 936641DEST_PATH_IMAGE002
, wherein
Figure 510842DEST_PATH_IMAGE003
Represent its equivalence on probability meaning;
Figure 647294DEST_PATH_IMAGE004
For zero-mean, variance are
Figure 736473DEST_PATH_IMAGE005
Gaussian random variable; zBe at random multiplier variable, then random vector YObey GSM and distribute its probability density function p(Y) be the mixing of a plurality of Gaussian distribution, and and multiplier
Figure 327991DEST_PATH_IMAGE006
Distribution
Figure 694382DEST_PATH_IMAGE007
Relevant, namely [11] [13]
Figure 64052DEST_PATH_IMAGE008
( )
Statistical property according to the NSS model, the edge of natural image wavelet conversion coefficient or joint probability distribution are obeyed GSM and are distributed, that is, if with in the same wavelet sub-band or adjacent yardstick or/and wavelet coefficient between directional subband elongates, can obtain obeying the random vector that GSM distributes Suppose a size to be
Figure 725475DEST_PATH_IMAGE010
(
Figure 734888DEST_PATH_IMAGE010
Enough little) the wavelet coefficient neighborhood in, multiplier
Figure 533080DEST_PATH_IMAGE006
Remain unchanged, so just can be temporarily with zRegard that one is determined amount or constant as.The center coefficient of remembering this neighborhood is
Figure 138505DEST_PATH_IMAGE011
, the random vector that all coefficients elongations obtain in the neighborhood is
Figure 276225DEST_PATH_IMAGE012
Subscript
Figure 987698DEST_PATH_IMAGE013
,
Figure 476448DEST_PATH_IMAGE014
Represent respectively center coefficient be positioned at the yardstick of wavelet coefficient subband ( Scale) and direction ( Orientation), subscript
Figure 10197DEST_PATH_IMAGE015
,
Figure 674528DEST_PATH_IMAGE016
Be illustrated respectively in row in the wavelet coefficient subband ( Row) and row ( Column) coordinate, therefore upper subscript has consisted of the index of center coefficient and neighborhood together.In the discussion of back, can not cause misunderstanding if omit upper subscript, then all no longer provide subscript.The multiplier that each neighborhood is corresponding
Figure 901110DEST_PATH_IMAGE017
(back is omitted and is expressed as
Figure 267369DEST_PATH_IMAGE006
) can pass through the conditional probability distribution function
Figure 604810DEST_PATH_IMAGE018
Maximal possibility estimation obtain.Do not losing universality ground hypothesis
Figure 982701DEST_PATH_IMAGE019
Prerequisite under, right
Figure 989972DEST_PATH_IMAGE006
The maximal possibility estimation result be [11]:
(
Figure 63287DEST_PATH_IMAGE021
)
Wherein
Figure 482636DEST_PATH_IMAGE022
Can be directly by original wavelet coefficient be estimated to obtain.
When to all wavelet coefficients, all in its corresponding coefficient neighborhood, estimate corresponding multiplier
Figure 926387DEST_PATH_IMAGE006
After, can obtain so the stochastic variable of approximate Gaussian distributed This process is called distinguishes normalization conversion (Divisive normalization transformation, DNT) process, each wavelet coefficient is by the energy institute normalization of the coefficient neighborhood at its place, thereby reduced the linear statistic correlation between the wavelet conversion coefficient, and with human eye vision mechanism in consistent to the distribution principle of energy [6] [11]The stochastic variable that obtains from the DNT process
Figure 290689DEST_PATH_IMAGE024
Be called as the DNT coefficient.
A large amount of correlative studys confirm natural image DNT coefficient
Figure 564544DEST_PATH_IMAGE024
Edge or joint probability distribution all be similar to Gaussian distributed, but when image fault, the distribution of its DNT coefficient begins to depart from Gaussian distribution, and the attenuation degree of departure degree and the human eye picture quality that can perceive is relevant [6] [10]Therefore, the distribution of DNT coefficient that can image just can be used in the Description Image whether distortion being arranged, and can by the DNT coefficient distribute and Gaussian distribution between the attenuation degree of extent of deviation quantitative estimation picture quality.
The optimum blind assessment algorithm of picture quality is a kind of blind assessment algorithm based on the NSS model that proposes in the document [6] at present, is called DIIVINE (Distortion Identification-based Image Verity and Integrity Evaluation) and estimates.DIIVINE estimates based on a double-layer frame realization, and the phase one is carried out type of distortion identification, and subordinate phase is estimated for the quality of carrying out of different distortions, at last two stage result integrated the quality estimated result that obtains image.But the DIIVINE algorithm needs complicated character selection and abstraction process.
Summary of the invention
The object of the invention is to the statistical property according to the blind assessment of picture quality, a kind of and the statistical measurement method of human eye subjective vision to the consistent blind assessment of picture quality of the assessment result of picture quality are provided.
The statistical measurement method of the blind assessment of picture quality that the present invention proposes, use distorted image and distinguish normalization conversion (Divisive Normalization Transform in wavelet field, DNT) statistical distribution of coefficient and its fit to the distortion that K-L statistical measurement between the Gaussian distribution is come Description Image, and concrete steps are as follows:
(1) if distorted image is coloured image, converts coloured image to gray level image first.Gray level image to distortion carries out the wavelet coefficient conversion, and the methods such as available steerable pyramid obtain
Figure 241513DEST_PATH_IMAGE025
Individual yardstick with
Figure 476186DEST_PATH_IMAGE026
Wavelet coefficient subband on the individual direction;
(2) in each wavelet coefficient subband, distinguish the normalization conversion, the multiplier variable of the Gauss's yardstick mixed distribution that namely the match wavelet coefficient is distributed according to formula (2) carries out maximal possibility estimation; With the multiplier estimated result of each wavelet coefficient divided by correspondence, obtain the DNT coefficient;
(3) in each wavelet coefficient subband, with the further piecemeal of DNT coefficient; In each DNT coefficient block, coefficient is gone average, the design factor variance; Variance to DNT coefficient block all in each coefficient subband is averaged, as the variance of the Gaussian distribution that is similar to the distribution of reference picture DNT coefficient;
Concrete operations are as follows: in each wavelet coefficient subband, Further Division is
Figure 303327DEST_PATH_IMAGE027
Individual size is
Figure 510318DEST_PATH_IMAGE028
Overlapped DNT coefficient block; With Individual DNT coefficient block pulls into the coefficient vector that obtains
Figure 80156DEST_PATH_IMAGE030
, the nothing of its variance is estimated as partially:
Figure DEST_PATH_IMAGE031
(
Figure 706396DEST_PATH_IMAGE032
)
With right The mean value of the variance that individual coefficient block estimation obtains
Figure DEST_PATH_IMAGE033
As the variance of the Gaussian distribution that is similar to the distribution of reference picture DNT coefficient, with undistorted variance
Figure 255506DEST_PATH_IMAGE034
Between have error
Figure 199191DEST_PATH_IMAGE035
, be shown below:
Figure 617403DEST_PATH_IMAGE036
(4)
Variance to DNT coefficient block all in each coefficient subband is averaged, as the variance of the Gaussian distribution that is similar to the distribution of reference picture DNT coefficient;
With zero-mean, variance be The Gaussian distribution DNT coefficient that is similar to undistorted image distribute DNT coefficient vector of distorted image so
Figure 785396DEST_PATH_IMAGE037
Will have deviation between the Gaussian distribution with the variance match that provides with zero-mean and formula (4) of distributing, the K-L that can use discretize has described the mass attenuation degree of the interior image of this coefficient subband apart from measurement;
(4) in each wavelet coefficient subband, calculate distribute K-L distance with the discretize of approximate Gaussian distribution of DNT coefficient, with the K-L that calculates apart from estimating as the mass attenuation of this wavelet coefficient subband;
Note DNT coefficient histogram vector is , subscript wherein dThe expression distortion ( Distorted),
Figure DEST_PATH_IMAGE039
The expression histogram index.The fitted Gaussian distribution vector is , subscript wherein fThe expression match ( Fitting); The DNT coefficient histogram of discretize and the distance of the K-L between the Gaussian distribution then are:
Figure 146474DEST_PATH_IMAGE041
(5)
(5) K-L of all wavelet coefficient subbands distance is averaged, estimates as the mass attenuation of the gray level image of this distortion, each coefficient subband is calculated
Figure 772627DEST_PATH_IMAGE042
, then to the estimated result of the quality of entire image decline be:
(6)
Wherein
Figure 936072DEST_PATH_IMAGE013
For the yardstick of wavelet coefficient subband ( Scale) index, For direction ( Orientation) index, For yardstick is Direction is
Figure 61843DEST_PATH_IMAGE014
The estimation of deviation result of coefficient subband.
According to the present invention, the statistical measurement value that every width of cloth image is obtained is larger, shows that then picture quality is poorer; The statistical measurement value is less, shows that then picture quality is better.
According to the present invention, directly from the wavelet coefficient of distorted image, estimate to can be used for the Gaussian Distribution Parameters as benchmark or reference, thereby can avoid the dependence to reference information, and can overcome character selection and abstraction that the blind assessment of conventional images quality relies on, machine learning etc.
According to the present invention, with DNT coefficient piecemeal, adopt the method for repeatedly measuring to reduce the error variance that Gaussian Distribution Parameters is estimated, thereby improve the accuracy of estimating.
The statistical measurement method of the blind assessment of picture quality that the present invention proposes, half reference evaluation method that proposes with document [10] similarly is, of the present invention estimate also rely on the priori statistical information that the NSS model is described natural image, and the DNT coefficient of distorted image distributes and Gaussian distribution between the deviation that exists.But different is: on the one hand, the present invention is directly from the wavelet coefficient of distorted image, estimation can be used for the Gaussian Distribution Parameters as benchmark or reference, thereby can avoid the dependence to reference information, and can overcome character selection and abstraction that the blind assessment of conventional images quality relies on, machine learning etc.; On the other hand, the present invention can to carry out parameter estimation much smaller than the Wavelet Coefficient Blocks of whole wavelet sub-band as unit, adopt the method for repeatedly measuring to reduce the error of estimating with reference to Gaussian Distribution Parameters.Be the present invention directly the DNT coefficient by distorted image distribution and come the approximate evaluation mass attenuation by the distance of the K-L between the Gaussian distribution of its match.Parameter as the Gaussian distribution of benchmark or reference is then directly estimated to obtain from the DNT coefficient of distorted image, and does not rely on the prior imformation etc. of any characteristic extraction procedure and reference picture, thereby can realize blind assessment truly.
Description of drawings
Fig. 1: DNT coefficient histogram example.The distorted image of the first behavior reference picture and the same different distortions of content, the image below be the subjective result (DMOS) of marking of human eye.The DNT coefficient histogram of a Wavelet Coefficient Blocks of the second behavior correspondence image and best-fit Gaussian distribution curve thereof, below are the K-L distance between DNT distribution and the fitted Gaussian distribution.
Fig. 2: the blind assessment result of objective image quality and subjective evaluation result contrast.
Embodiment
Provided the DNT coefficient histogram example of some reference pictures and distorted image among Fig. 1.The first row is that a width of cloth reference picture and several have identical content, but has the distorted image of different distortion levels, and image is from the LIVE database [14], every width of cloth image below has provided the subjective marking of the human eye result (DMOS) who comprises among the corresponding distortion class of this image and the LIVE.The scope of DMOS is [0,100], and the DMOS=0 representative image is undistorted, and along with the distortion level increase of image, the value of DMOS also can correspondingly increase.The second row is DNT coefficient histogram and the best-fit Gaussian curve thereof of a wavelet coefficient subband of image, and the below has provided the K-L distance that estimation obtains, i.e. deviation between DNT coefficient distribution and the Gaussian distribution.Deviate is larger, and the representative image mass attenuation is more serious.
From Fig. 1, can see intuitively the DNT coefficient distribution of image and the relation between the picture quality.Consistent with the NSS model property, the DNT coefficient of reference picture distributes and is similar to Gaussian distributed; And when introducing distortion in the image, the DNT coefficient distributes and begins to depart from Gaussian distribution, and departure degree is relevant with the attenuation degree of picture quality, and the K-L distance is also along with the picture quality decling phase should increase.
This experiment is at LIVE image quality measure database [14]Upper checking is of the present invention estimates performance.Selecting the reason of LIVE database is wherein to have comprised five kinds of different distortion class images: JP2K, JPEG, white noise, Gaussian Blur and channel fast-fading, and provided subjective evaluation mark (DMOS) corresponding to every width of cloth image.In addition, the LIVE database in a lot of image quality measure pertinent literatures as benchmark database test pattern method for evaluating quality performance [1] [3-6] [10]
Concrete experimental technique is: all use blind assessment statistical measurement of the present invention to obtain a picture quality estimated score to the every width of cloth image in the LIVE database, then the subjective evaluation mark (DMOS) that provides in the database and the estimated result that blind assessment statistical measurement obtains are done the scatter diagram curve.Provided the matched curve of correspondence of scatter diagram on different distortion class subdata bases and whole database among Fig. 2.
As seen from Figure 2: to JPEG, Gaussian Blur and channel fast-fading distortion class, loose point relatively concentrates near the matched curve.But dialogue is made white noise distortion class, and the distribution of loose point then relatively disperses.In the scatter diagram that whole LIVE database is obtained, the result of statistical measurement assessment of the present invention relatively concentrates near the matched curve, shows that its total evaluation result is reliable.
Quantitatively weigh validity of the present invention with the estimated result of blind assessment statistical measurement of the present invention and five objective evaluation indexs between the matched curve: 1) linearly dependent coefficient (Linear Correlation Coefficient, LCC), described the accuracy of prediction, value is more higher close to 1 expression forecasting accuracy; 2) absolute error (Medium Absolute Error, MAE) is worth less expression and predicts that absolute error is less; 3) square error root (Root Mean Squared Error, RMSE) is worth less expression and predicts that square error is less; 4) unusual ratio (Outlier Ratio, OR) has been described the prediction consistance, is worth less expression and predicts that consistance is higher; 5) Spearman rank correlation coefficient (Spearman ' s Rank Ordered Correlation Coefficient, SROCC) is described the monotonicity of prediction, and value is more more excellent close to 1 expression prediction monotonicity.And with effect of the present invention and two kinds of full reference picture quality evaluation indexs---PSNR and SSIM [4], a kind of blind appraisal procedure---DIIVINE [6]Compare.The reason of selecting these methods to compare be because: PSNR is the complete in assessment algorithm of widespread use with the longest history and, and SSIM is be subject at present extensively approval complete in assessment algorithm, and DIIVINE be at present known to optimum blind assessment algorithm.Obtain following result:
Table
Figure 806945DEST_PATH_IMAGE009
: the different Performance Ratios of image quality measure algorithm on the LIVE database are.Italic be blind assessment algorithm, other is for entirely with reference to assessment algorithm.
The objective indicator result who provides from table 1 can find out, statistical measurement of the present invention can provide with human eye subjective to the higher assessment result of minute consistance, and overall performance is better than the blind assessment algorithm DIIVINE of present optimum known to us [6]DIIVINE algorithm performance index is to use from the source code of [15] download to test at LIVE, obtains by loose point curve match again.Statistical measurement of the present invention also needs complicated feature extraction and learning process unlike DIIVINE, when having low computation complexity, the assessment result of most distortion classes and whole database all is better than the DIIVINE algorithm.
List of references
[1] H. R. Sheikh, M. F. Sabir, and A. C. Bovik. A Statistical Evaluation of Recent Full Reference Image Quality Assessment Algorithms [J]. IEEE Trans. Image Process., 2006, 15(11):3441-3452.
[2] Z. Wang. Applications of Objective Image Quality Assessment Methods [J]. IEEE Signal Process. Magazine, 2011, 15(2):137-142.
[3] H. R. Sheikh and A. C. Bovik. Image information and visual quality [J]. IEEE Trans. Image Process., 2006, 15(2):430-444.
[4] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli. Image quality assessment: From error measurement to structural similarity [J]. IEEE Trans. Image Process., 2006, 13(4):600-612.
[5] M. A. Saad, A. C. Bovik, and C. Charrier. DCT statistics model-based blind image quality assessment [J]. IEEE Int. Conf. Image Processing, Brussels, Belgium, 2010.
[6] A. K. Moorthy and A. C. Bovik. Blind image quality assessment: from natural scene statistics to perceptual quality [J]. IEEE Tran. Image Process., 2011, 20(12): 3350-3363.
[7] A. Mittal, G. S. Muralidhar, J. Ghosh, and A. C. Bovik. Blind image quality assessment without human training using latent quality factors [J]. IEEE Tran. Signal Process., 2011, 19(2):75-78.
[8] S. Gabarda and G. Cristobal. Blind image quality assessment through anisotropy [J]. J. Optic. Soc. Amer. A, 2007, 24(12):42-51.
[9] Z. Wang and A. C. Bovik. Reduced- and no-reference image quality assessment [J]. IEEE Signal Process. Magazine, 2011, 28(6):29-40.
[10] Q. Li, Z. Wang. Reduced-reference image quality assessment using divisive normalization-based image representation [J]. IEEE J. Select. Topics Signal Process., 2009, 3(2):202-211.
[11] M. J. Wainwright and E. P. Simoncelli. Scale mixtures of Gaussians and the statistics of natural images [J]. Adv. Neural Inf. Process. Syst., 2000, 12(1): 855–861.
[12] J. Portilla, V. Strela, M. J. Wainwright, and E. P. Simoncelli. Image denoising using scale mixture of Gaussians in the wavelet domain [J]. IEEE Trans. Image Process., 2003, 12(11):1338-1351.
[13] A. K. Moorthy and A. C. Bovik. Statistics of natural image distortions [J]. Proc. IEEE Int. Conf, Acoustics Speech and Signal Processing (ICASSP), 2010, 962-965.
[14] H. R. Sheikh, Z. Wang, A. C. Bovik, and L. K. Cormack, “Image and Video Quality Assessment Research at LIVE.” [Online]. Available: http://live.ece.utexas.edu/research/quality
[15] A. K. Moorthy and A. C. Bovik, [Online]. Available: http://live.ece.utexas.edu/research /quality/DIIVINE_release.zip。

Claims (4)

1. the statistical measurement method of a blind assessment of picture quality, it is characterized in that using distorted image and distinguish the distortion that K-L statistical measurement between the Gaussian distribution of the statistical distribution of normalization conversion (DNT) coefficient and its match is come Description Image in wavelet field, concrete steps are as follows:
(1) if distorted image is coloured image, converts coloured image to gray level image first; Gray level image to distortion carries out the wavelet coefficient conversion, obtains Individual yardstick with
Figure DEST_PATH_IMAGE004
Wavelet coefficient subband on the individual direction;
(2) in each wavelet coefficient subband, distinguish the normalization conversion, the multiplier variable of the Gauss's yardstick mixed distribution that namely the match wavelet coefficient is distributed carries out maximal possibility estimation; With the multiplier estimated result of each wavelet coefficient divided by correspondence, obtain the DNT coefficient;
(3) in each wavelet coefficient subband, with the further piecemeal of DNT coefficient; In each DNT coefficient block, coefficient is gone average, the design factor variance; Variance to DNT coefficient block all in each coefficient subband is averaged, as the variance of the Gaussian distribution that is similar to the distribution of reference picture DNT coefficient;
(4) in each wavelet coefficient subband, calculate distribute K-L distance with the discretize of approximate Gaussian distribution of DNT coefficient, with the K-L that calculates apart from estimating as the mass attenuation of this wavelet coefficient subband;
(5) K-L of all wavelet coefficient subbands distance is averaged, estimates as the mass attenuation of the gray level image of this distortion.
2. method according to claim 1 is characterized in that the statistical measurement value that every width of cloth image is obtained is larger, shows that then picture quality is poorer; The statistical measurement value is less, shows that then picture quality is better.
3. method according to claim 1, it is characterized in that directly from the wavelet coefficient of distorted image, estimation can be used for the Gaussian Distribution Parameters as benchmark or reference, avoiding the dependence to reference information, and overcome character selection and abstraction, machine learning that the blind assessment of conventional images quality relies on.
4. method according to claim 1 is characterized in that DNT coefficient piecemeal, adopts the method for repeatedly measuring to reduce the error variance that Gaussian Distribution Parameters is estimated, to improve the accuracy of estimating.
CN2012104406180A 2012-11-07 2012-11-07 Statistical measure method for image quality blind estimation Pending CN102930545A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2012104406180A CN102930545A (en) 2012-11-07 2012-11-07 Statistical measure method for image quality blind estimation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2012104406180A CN102930545A (en) 2012-11-07 2012-11-07 Statistical measure method for image quality blind estimation

Publications (1)

Publication Number Publication Date
CN102930545A true CN102930545A (en) 2013-02-13

Family

ID=47645335

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2012104406180A Pending CN102930545A (en) 2012-11-07 2012-11-07 Statistical measure method for image quality blind estimation

Country Status (1)

Country Link
CN (1) CN102930545A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103325113A (en) * 2013-06-06 2013-09-25 深圳大学 Method and device for reduced-reference image quality assessment
CN103366378A (en) * 2013-07-26 2013-10-23 深圳大学 Reference-free type image quality evaluation method based on shape consistency of condition histogram
CN104268590A (en) * 2014-09-17 2015-01-07 电子科技大学 Blind image quality evaluation method based on complementarity combination characteristics and multiphase regression
CN104282019A (en) * 2014-09-16 2015-01-14 电子科技大学 Blind image quality evaluation method based on natural scene statistics and perceived quality propagation
CN107666472A (en) * 2016-07-29 2018-02-06 微软技术许可有限责任公司 The digital simulation encoding and decoding of mixing
CN110363763A (en) * 2019-07-23 2019-10-22 上饶师范学院 Image quality evaluating method, device, electronic equipment and readable storage medium storing program for executing

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101977333A (en) * 2010-11-24 2011-02-16 南京信息工程大学 Non-reference image quality evaluating method based on wavelet and structural self-similarity analysis
US20110038548A1 (en) * 2009-02-11 2011-02-17 Rezazadeh Soroosh Method and system for determining a quality measure for an image using multi-level decomposition of images

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110038548A1 (en) * 2009-02-11 2011-02-17 Rezazadeh Soroosh Method and system for determining a quality measure for an image using multi-level decomposition of images
CN101977333A (en) * 2010-11-24 2011-02-16 南京信息工程大学 Non-reference image quality evaluating method based on wavelet and structural self-similarity analysis

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ANUSH KRISHNA MOORTHY ET AL: "《Blind Image Quality Assessment:From Natural Scene Statistics to Perceptual Quality》", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *
MARTIN J.WAINWRIGHT ET AL: "《Scale Mixtures of Gaussians and the Statistics of Nature Images》", 《ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 12,》 *
QIANG LI ET AL: "《Reduced-Reference Image Quality Assessment Using Divisive Normalization-Based Image Representation》", 《IEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING》 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103325113A (en) * 2013-06-06 2013-09-25 深圳大学 Method and device for reduced-reference image quality assessment
CN103325113B (en) * 2013-06-06 2016-03-09 深圳大学 Partial reference type image quality evaluating method and device
CN103366378A (en) * 2013-07-26 2013-10-23 深圳大学 Reference-free type image quality evaluation method based on shape consistency of condition histogram
CN103366378B (en) * 2013-07-26 2016-01-20 深圳大学 Based on the no-reference image quality evaluation method of conditional histograms shape coincidence
CN104282019A (en) * 2014-09-16 2015-01-14 电子科技大学 Blind image quality evaluation method based on natural scene statistics and perceived quality propagation
CN104282019B (en) * 2014-09-16 2017-06-13 电子科技大学 Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated
CN104268590A (en) * 2014-09-17 2015-01-07 电子科技大学 Blind image quality evaluation method based on complementarity combination characteristics and multiphase regression
CN104268590B (en) * 2014-09-17 2017-08-11 电子科技大学 The blind image quality evaluating method returned based on complementary combination feature and multiphase
CN107666472A (en) * 2016-07-29 2018-02-06 微软技术许可有限责任公司 The digital simulation encoding and decoding of mixing
CN107666472B (en) * 2016-07-29 2020-08-11 微软技术许可有限责任公司 Method and apparatus for hybrid digital-analog coding
US10869029B2 (en) 2016-07-29 2020-12-15 Microsoft Technology Licensing, Llc Hybrid digital-analog coding
CN110363763A (en) * 2019-07-23 2019-10-22 上饶师范学院 Image quality evaluating method, device, electronic equipment and readable storage medium storing program for executing

Similar Documents

Publication Publication Date Title
Wang et al. Reduced-reference image quality assessment using a wavelet-domain natural image statistic model
Feichtenhofer et al. A perceptual image sharpness metric based on local edge gradient analysis
Wang et al. Reduced-and no-reference image quality assessment
Saad et al. DCT statistics model-based blind image quality assessment
CN103475898B (en) Non-reference image quality assessment method based on information entropy characters
Liang et al. No-reference perceptual image quality metric using gradient profiles for JPEG2000
CN101562675B (en) No-reference image quality evaluation method based on Contourlet transform
Guan et al. No-reference blur assessment based on edge modeling
CN102930545A (en) Statistical measure method for image quality blind estimation
CN101877127B (en) Image reference-free quality evaluation method and system based on gradient profile
Wang et al. Novel spatio-temporal structural information based video quality metric
CN105825500A (en) Camera image quality evaluation method and device
CN109978854B (en) Screen content image quality evaluation method based on edge and structural features
Kim et al. Gradient information-based image quality metric
Zheng et al. No-reference quality assessment for screen content images based on hybrid region features fusion
CN101976444B (en) Pixel type based objective assessment method of image quality by utilizing structural similarity
CN103258326B (en) A kind of information fidelity method of image quality blind evaluation
CN104202594B (en) A kind of method for evaluating video quality based on 3 D wavelet transformation
CN101989355A (en) Detection method for secondary compression of JPEG image
Ni et al. Screen content image quality assessment using edge model
CN102036098B (en) Full-reference type image quality evaluation method based on visual information amount difference
Zhang et al. Reduced reference image quality assessment based on statistics of edge
Tang et al. Training-free referenceless camera image blur assessment via hypercomplex singular value decomposition
Liang et al. A no-reference perceptual blur metric using histogram of gradient profile sharpness
CN103996188A (en) Full-reference-type image quality evaluation method based on Gabor weighted characteristics

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20130213