CN104282019B - Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated - Google Patents

Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated Download PDF

Info

Publication number
CN104282019B
CN104282019B CN201410473339.3A CN201410473339A CN104282019B CN 104282019 B CN104282019 B CN 104282019B CN 201410473339 A CN201410473339 A CN 201410473339A CN 104282019 B CN104282019 B CN 104282019B
Authority
CN
China
Prior art keywords
image
quality
marked
test image
feature
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.)
Expired - Fee Related
Application number
CN201410473339.3A
Other languages
Chinese (zh)
Other versions
CN104282019A (en
Inventor
李宏亮
吴庆波
熊健
李威
罗冰
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.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201410473339.3A priority Critical patent/CN104282019B/en
Publication of CN104282019A publication Critical patent/CN104282019A/en
Application granted granted Critical
Publication of CN104282019B publication Critical patent/CN104282019B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30168—Image quality inspection

Landscapes

  • Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Testing, Inspecting, Measuring Of Stereoscopic Televisions And Televisions (AREA)
  • Image Analysis (AREA)

Abstract

The present invention provides a kind of blind image quality evaluating method propagated based on natural scene statistics and perceived quality, expert FoE gradient responses are calculated to test image and a large amount of undistorted natural images, statistics response histogram distribution, the KL divergences of both distributions are calculated, the absolute distortion level of test image is obtained.Quality Perception Features are extracted to test image and marked distorted image, and according to the card side's distance between feature, finds the N number of mark image most like with test image.By the way that these to be marked the quality score weighted sum of image, the relative distortion level of test image is can obtain.Finally, predict that marking obtains final prognostic chart picture mass fraction by being grouped together by first two steps.Compared to existing representative non-reference picture quality appraisement method, the method is simply efficient, and without a large amount of handmarking's samples.

Description

Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated
Technical field
The present invention relates to image processing techniques, visual signal treatment technology is more particularly to perceived.
Background technology
Efficient image perception quality evaluating method is then the key technology in multimedia service quality monitoring field.At present, Reliable image quality evaluating method is mainly full reference and weak reference type.These methods are required to access nothing completely The artwork information of distortion.However, in the middle of many applied environments, this requirement cannot often meet.
The information that blind image (non-reference picture) quality evaluating method only needs to distorted image itself can be predicted its perception Quality.Existing blind image quality evaluating method is returned etc. often through supporting vector the learning method of supervision directly training figure As the black box projection model that feature and perceived quality are given a mark is used for image quality estimation.In order to ensure the robustness of model, these Method needs substantial amounts of handmarking's image for training.Simultaneously as the characteristics of its black box is projected, these methods cannot be clear Relation between description characteristics of image and perceived quality.
The content of the invention
The technical problems to be solved by the invention are to provide a kind of relation that can be described between characteristics of image and quality score Blind reference image quality appraisement method.
The present invention is to solve the technical scheme that above-mentioned technical problem is used, based on natural scene statistics and perceived quality The blind image quality evaluating method propagated, comprises the following steps：
Step 1) natural scene statistics：
Expert FoE (Fields of Experts) gradient response of test image and all undistorted images is calculated, And statistics obtains the FoE gradient response histogram distributions P of test image respectivelydAnd i-th FoE response of undistorted image Value histogram is distributed PuI (), i=1,2 ..., K, K are undistorted image sum, calculate the phase of test image and undistorted image To entropy KL divergences
Step 2) perceived quality propagation：
1-1：Image texture characteristic is extracted to test image, global Gradient Features and the boundary intensity based on down-sampling are special Levy；The extracting method of the boundary intensity feature of the down-sampling is：1/8 down-sampling is carried out to image, after down-sampling image each The boundary strength value of point represents by its maximum for vertically and horizontally going up gradient, then to boundary strength value a little carry out Statistics with histogram, the histogram after normalization is the boundary intensity feature based on down-sampling；
1-2：Test image is calculated with the feature cards side of marked distorted image apart from D, the marked distortion map As being the distorted image for having carried out picture quality marking by way of handmarking；WhereinWithThe ith feature vector of vectorial, the marked distorted image of the ith feature of test image, i={ 1,2,3 } are represented respectively Difference correspondence image textural characteristics, global Gradient Features and the boundary intensity feature based on down-sampling；
1-3：The marked distorted image of sequential selection top n according to feature cards side apart from D from small to large；And according to this N Individual feature cards side calculates respective weight w apart from Dn,DnRepresent test image with order from small to large Feature cards side's distance of n-th marked distorted image of selection；
1-4：Using weight wnTo the image quality score DMOS of N number of marked distorted imagenSummation is weighted to be surveyed Attempt the prediction fraction Q of picturePQP,
Step 3) relative entropy KL divergences Q is setNSSAnd prediction fraction QPQPWeight parameter, with reference to relative entropy KL divergences QNSSAnd prediction fraction QPQPObtain final prediction marking Q.
The present invention calculates expert FoE gradient responses to test image and a large amount of undistorted natural images, unites respectively Count the response histogram distribution of test image and all undistorted images.Again by calculating the KL divergences of both distributions, we The absolute distortion level of test image can be obtained.Secondly, we extract quality to test image and marked distorted image Perception Features, and according to the card side's distance between feature, find the N number of mark image most like with test image.By by these The quality score weighted sum of image is marked, the relative distortion level of test image is can obtain.Finally, predicted by first two steps Marking obtains final prognostic chart picture mass fraction by being grouped together.
The beneficial effects of the invention are as follows compared with existing representative non-reference picture quality appraisement method, the party Method is simply efficient, and without a large amount of handmarking's samples.
Brief description of the drawings
Fig. 1：Block schematic illustration of the present invention.
Specific embodiment
Effectively to carry out quality evaluation to non-reference picture, the present invention is made up of three steps：Natural scene statistics step Suddenly, perceived quality propagation steps, comprehensive marking step.Wherein, natural scene statistics is by comparing test image with a large amount of without mistake The statistical discrepancy of true natural image obtains the assessment of absolute distortion information.And mass propagation marks image by by part Quality score is broadcast to similar test image, can obtain the assessment of relative distortion information.Finally, by by two modules Prediction marking combines is given a mark with obtaining final prediction.
The present embodiment is realized on matlab2009b software platforms, specific as shown in Figure 1：
Step one, the FoE gradient responses for calculating test image and selected undistorted image, and count both respectively Histogram distribution.Allow PdRepresent the response distribution of test image, PuRepresent the response distribution of whole undistorted images.Then natural field The quality score of scape statistical module is represented by both KL divergences, i.e.,
FoE gradients response distribution with undistorted image if test image distortion is had any different, when test image is fuzzy, Then the FoE gradients response distribution of test image is flat compared with undistorted image, and when test image has noise, then its FoE gradient is rung Should be distributed and folder peak occurs.
Step 2, perceived quality are propagated and are mainly made up of following three step：
1st step：The feature of mass-sensitive, including image texture characteristic, global Gradient Features and base are extracted to test image In the boundary intensity DSBS features of down-sampling, wherein image texture characteristic can be by SFTA (Segmentation-based Fractal Texture Analysis) feature instantiation, global Gradient Features are by GIST feature instantiations.For based on down-sampling Boundary intensity feature extraction, 1/8 down-sampling is carried out to image first, after sampling image each point boundary intensity hung down by it The maximum of gradient is represented in straight and horizontal direction.Then, to boundary strength value a little carry out statistics with histogram, normalize Histogram afterwards is DSBS characteristic vectors, and DSBS features are used for the blocking effect reflected after compression of images.
2nd step：AllowWithThe ith feature vector of test image and marked image is represented,Represent two Card side's distance of person, then test image and the total characteristic distance of mark image are represented byWherein i= { 1,2,3 } corresponds to SFTA, GIST and DSBS features respectively.
3rd step：5 marked images for making D minimum are found, and respective weights are calculated according to their characteristic distance wn：
DnRepresent n-th marked distortion map of test image and sequential selection from small to large Feature cards side's distance of picture；
Then, the prediction marking of the module is represented by the weighted sum of the mass fraction DMOS of marked image, here The span of mass fraction DMOS be 0 to 100,0 represent it is best, 100 represent it is worst：
Step 3, combined by by first two steps quality score, you can obtain final prediction marking
Wherein γ is two weight parameters of module, and here, we are set to 0.2.

Claims (4)

1. the blind image quality evaluating method propagated based on natural scene statistics and perceived quality, it is characterised in that including following Step：
Step 1) natural scene statistics：
The expert FoE gradient responses of test image and all undistorted images are calculated, and statistics obtains test image respectively FoE gradient response histogram distributions PdAnd i-th FoE response histogram distributions P of undistorted imageu(i), i=1, 2 ..., K, K are undistorted image sum, calculate the relative entropy KL divergences of test image and undistorted image
Q NSS = Σ i = 1 K P d log 2 ( P d P u ( i ) ) ;
Step 2) perceived quality propagation：
1-1：Image texture characteristic, global Gradient Features and the boundary intensity feature based on down-sampling are extracted to test image；Institute The extracting method for stating the boundary intensity feature of down-sampling is：1/8 down-sampling is carried out to image, the side of each point of image after down-sampling Boundary's intensity level represents by its maximum for vertically and horizontally going up gradient, then to boundary strength value a little enter column hisgram Statistics, the histogram after normalization is the boundary intensity feature based on down-sampling；
1-2：Test image is calculated with the feature cards side of marked distorted image apart from D, the marked distorted image is The distorted image of picture quality marking has been carried out by way of handmarking；Wherein Fi qAnd Fi rPoint Not Biao Shi test image vectorial, the marked distorted image of ith feature ith feature vector, i={ 1,2,3 } is respectively Correspondence image textural characteristics, global Gradient Features and the boundary intensity feature based on down-sampling；
1-3：The marked distorted image of sequential selection top n according to feature cards side apart from D from small to large；And according to this N number of spy Levy card side and calculate respective weight w apart from Dn,DnRepresent test image with sequential selection from small to large N-th marked distorted image feature cards side's distance；
1-4：Using weight wnTo the image quality score DMOS of N number of marked distorted imagenIt is weighted summation and obtains test chart The prediction fraction Q of picturePQP,
Step 3) relative entropy KL divergences Q is setNSSAnd prediction fraction QPQPWeight parameter, with reference to relative entropy KL divergences QNSSWith And prediction fraction QPQPObtain final prediction marking Q.
2. the blind image quality evaluating method propagated based on natural scene statistics and perceived quality as claimed in claim 1, it is special Levy and be, the final prediction marking Q isγ is weight parameter.
3. the blind image quality evaluating method propagated based on natural scene statistics and perceived quality as claimed in claim 2, it is special Levy and be, γ=0.2.
4. the blind image quality evaluating method propagated based on natural scene statistics and perceived quality as claimed in claim 1, it is special Levy and be, N=5.
CN201410473339.3A 2014-09-16 2014-09-16 Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated Expired - Fee Related CN104282019B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410473339.3A CN104282019B (en) 2014-09-16 2014-09-16 Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410473339.3A CN104282019B (en) 2014-09-16 2014-09-16 Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated

Publications (2)

Publication Number Publication Date
CN104282019A CN104282019A (en) 2015-01-14
CN104282019B true CN104282019B (en) 2017-06-13

Family

ID=52256869

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410473339.3A Expired - Fee Related CN104282019B (en) 2014-09-16 2014-09-16 Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated

Country Status (1)

Country Link
CN (1) CN104282019B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104902277B (en) * 2015-06-08 2018-03-09 浙江科技学院 One kind is based on singly drilling binary-coded non-reference picture quality appraisement method
CN106910180B (en) * 2015-12-22 2019-08-20 成都理想境界科技有限公司 A kind of image quality measure method and device
CN106815839B (en) * 2017-01-18 2019-11-15 中国科学院上海高等研究院 A kind of image quality blind evaluation method
CN109635142B (en) * 2018-11-15 2022-05-03 北京市商汤科技开发有限公司 Image selection method and device, electronic equipment and storage medium
CN109584242A (en) * 2018-11-24 2019-04-05 天津大学 Maximum entropy and KL divergence are without reference contrast distorted image quality evaluating method
CN111932521B (en) * 2020-08-13 2023-01-03 Oppo（重庆）智能科技有限公司 Image quality testing method and device, server and computer readable storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102930545A (en) * 2012-11-07 2013-02-13 复旦大学 Statistical measure method for image quality blind estimation
CN103258326A (en) * 2013-04-19 2013-08-21 复旦大学 Information fidelity method for image quality blind evaluation
CN103778636A (en) * 2014-01-22 2014-05-07 上海交通大学 Feature construction method for non-reference image quality evaluation

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100955180B1 (en) * 2008-04-11 2010-04-30 엔에이치엔(주) Method and System for Computing Quality Value of Image

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102930545A (en) * 2012-11-07 2013-02-13 复旦大学 Statistical measure method for image quality blind estimation
CN103258326A (en) * 2013-04-19 2013-08-21 复旦大学 Information fidelity method for image quality blind evaluation
CN103778636A (en) * 2014-01-22 2014-05-07 上海交通大学 Feature construction method for non-reference image quality evaluation

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Blind Image Quality Assessment: From Natural Scene Statistics to Perceptual Quality;Anush Krishna Moorthy, et al.;《IEEE TRANSACTIONS ON IMAGE PROCESSING》;20111231;第20卷(第12期);第3350-3364页 *
No Reference Image Quality Metric via Distortion Identification and Multi-Channel Label Transfer;Qingbo Wu, et al.;《2014 IEEE International Symposium on Circuits and Systems (ISCAS)》;20140605;第530-533页 *

Also Published As

Publication number Publication date
CN104282019A (en) 2015-01-14

Similar Documents

Publication Publication Date Title
CN104282019B (en) Based on the blind image quality evaluating method that natural scene statistics and perceived quality are propagated
CN103996192B (en) Non-reference image quality evaluation method based on high-quality natural image statistical magnitude model
CN102333233B (en) Stereo image quality objective evaluation method based on visual perception
CN103164711B (en) The method of region based on pixel and support vector machine artificial abortion's density Estimation
CN106447646A (en) Quality blind evaluation method for unmanned aerial vehicle image
CN102945552A (en) No-reference image quality evaluation method based on sparse representation in natural scene statistics
CN110033446A (en) Enhancing image quality evaluating method based on twin network
CN104867138A (en) Principal component analysis (PCA) and genetic algorithm (GA)-extreme learning machine (ELM)-based three-dimensional image quality objective evaluation method
CN104574381B (en) A kind of full reference image quality appraisement method based on local binary patterns
CN105338343A (en) No-reference stereo image quality evaluation method based on binocular perception
CN106127741A (en) Non-reference picture quality appraisement method based on improvement natural scene statistical model
CN102547368A (en) Objective evaluation method for quality of stereo images
CN107396095A (en) One kind is without with reference to three-dimensional image quality evaluation method
CN106506901A (en) A kind of hybrid digital picture halftoning method of significance visual attention model
CN102663747A (en) Stereo image objectivity quality evaluation method based on visual perception
CN104616294A (en) Visual saliency-based tone mapping image quality evaluation method
CN104268590A (en) Blind image quality evaluation method based on complementarity combination characteristics and multiphase regression
CN104954778A (en) Objective stereo image quality assessment method based on perception feature set
CN104881867A (en) Method for evaluating quality of remote sensing image based on character distribution
CN104021567B (en) Based on the fuzzy altering detecting method of image Gauss of first numeral law
CN103745466A (en) Image quality evaluation method based on independent component analysis
CN108550146A (en) A kind of image quality evaluating method based on ROI
CN104866864A (en) Extreme learning machine for three-dimensional image quality objective evaluation
CN106339677A (en) Video-based railway wagon dropped object automatic detection method
CN103093243A (en) High resolution panchromatic remote sensing image cloud discriminating method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20170613

Termination date: 20190916