CN109919894A - A kind of non-reference picture quality appraisement method and system based on human visual system - Google Patents
A kind of non-reference picture quality appraisement method and system based on human visual system Download PDFInfo
- Publication number
- CN109919894A CN109919894A CN201711286394.1A CN201711286394A CN109919894A CN 109919894 A CN109919894 A CN 109919894A CN 201711286394 A CN201711286394 A CN 201711286394A CN 109919894 A CN109919894 A CN 109919894A
- Authority
- CN
- China
- Prior art keywords
- image
- gray level
- feature vector
- occurrence matrixes
- mapping relations
- 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
Links
Landscapes
- Testing, Inspecting, Measuring Of Stereoscopic Televisions And Televisions (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of non-reference picture quality appraisement method and system based on human visual system, the method calculates its corresponding characteristic image according to training image, it calculates the gray level co-occurrence matrixes of each characteristic image and calculates separately the feature vector of each gray level co-occurrence matrixes, establish feature vector to image quality score mapping relations model, the corresponding gray level co-occurrence matrixes feature vector of testing image is brought into the mapping relations model, the mass fraction of testing image is obtained；The system comprises model generation module and prediction of quality modules；The model generation module is used to calculate the gray level co-occurrence matrixes of characteristic image, and establishes mapping relations model according to the feature vector of gray level co-occurrence matrixes；The prediction of quality module is used to obtain the mass fraction of testing image according to the gray level co-occurrence matrixes feature vector and the mapping relations model of testing image；Described method and system has reached and the very high consistency of human visual system.
Description
Technical field
The present invention relates to field of image processings, more particularly, to a kind of non-reference picture based on human visual system
Quality evaluating method and system.
Background technique
With the development of scientific progress, image processing techniques and a theoretical important neck for having become computer application
Domain is widely used in numerous science and engineering, and image quality evaluation is particularly significant in field of image processing, it is optimization figure
As the important performance indexes of processing system；Image quality evaluation can be divided into subjective assessment and objectively evaluate, and subjective assessment relies on people
For visual cognition impression evaluation image quality, the evaluation result of subjective assessment be more in line with human eye cognition impression, but because
It is very big for the uncertain factor of people, so subjective image quality evaluation reliability is poor, and objective image processing evaluation basis
Available reference image information number, figure can be divided into: full reference mass evaluation method, part reference mass evaluation method
And reference-free quality evaluation method, the reference refer to using the complete information of original image as the reference of evaluation；And it is most
In number situation, the information of original image can not be often obtained, thus in the practice of objective image processing evaluation, no ginseng
Examining quality evaluating method has very high application value；Profile, texture according to the subjective assessment of visual cognition impression to image
Feature it is very sensitive, the evaluation knot that the evaluation result obtained in reference-free quality evaluation method is often obtained with subjective assessment
Fruit has differences in varying degrees, and the consistency of reference-free quality evaluation method and human visual system are still poor at present.
Summary of the invention
In order to solve the consistency of current reference-free quality evaluation method and human visual system existing for background technique still
Right poor, the present invention provides a kind of non-reference picture quality appraisement method and system based on human visual system, institute
State method and system be utilized different interval distance gray level co-occurrence matrixes simulate human eye viewing distance variation influence, use
Gradient image and wavelet transformation solve the sensibility capture to image outline and textural characteristics；It is described a kind of based on human vision
The non-reference picture quality appraisement method of characteristic includes:
Step 1, its corresponding characteristic image is calculated according to training image, the characteristic image include training image and its
Corresponding gradient image, horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image；
Step 2, the gray level co-occurrence matrixes of each characteristic image are calculated；
Step 3, the feature vector of each gray level co-occurrence matrixes is calculated separately；
Step 4, establish feature vector to image quality score mapping relations model；
Step 5, the corresponding gray level co-occurrence matrixes feature vector of testing image is brought into the mapping relations model, is obtained
The mass fraction of testing image；
Further, the calculation method for calculating its corresponding gradient image according to distorted image is
Wherein, G is gradient image, GxFor the gradient component of horizontal direction, GyFor the gradient component of vertical direction；The Gx
It is obtained by gradient template and distorted image convolution, the GyIt is obtained by the transposition and distorted image convolution of gradient template；Institute
Stating gradient template is [- 10 1]；
Further, the wavelet transformation that 1 scale is carried out to the training image obtains horizontal high frequency component image, vertical
High fdrequency component image and oblique high fdrequency component image；
Further, the image gray levels of the characteristic image are quantified as L grades, gray level i and gray level j folder after quantization
Angle is θ, and spacing distance is d between two pixels, then the gray level co-occurrence matrixes are expressed as [P (i, j, d, θ)]L*L；
Wherein, 1≤i≤L, 1≤j≤L；I, j and L is natural number；0 °≤θ≤360 °, the value quantity of θ is M；d
Value quantity be it is N number of, d is positive number；The then characteristic image quantity number that the quantity of the gray level co-occurrence matrixes is M*N times；
Further, described eigenvector includes correlation, comparative, energy and local entropy；
Further, using support vector regression establish feature vector to image quality score mapping relations model；
Further, the mapping relations model of described eigenvector to image quality score is repeated according to multiple training images
The multiple groups feature vector that step 1 is obtained to 3 is established；
A kind of non-reference picture quality appraisement system based on human visual system includes:
Model generation module and prediction of quality module；
The model generation module is used to calculate the gray level co-occurrence matrixes of characteristic image, and according to each gray level co-occurrence matrixes
Feature vector establish feature vector to image quality score mapping relations model；The characteristic image include training image with
And the corresponding gradient image of training image, horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component figure
Picture；The mapping relations model that feature vector establishes feature vector to image quality score is sent to matter by the model generation module
Measure prediction module；
The prediction of quality module is used to be closed according to the gray level co-occurrence matrixes feature vector of testing image and the mapping
It is the mass fraction that model obtains testing image；
Further, the model generation module establishes feature vector to image quality score using support vector regression
Mapping relations model；
Further, the training image is one or more；Characteristic image described in each training image is one group corresponding；
The invention has the benefit that technical solution of the present invention, give it is a kind of based on human visual system without ginseng
Image quality evaluating method and system are examined, the method and system are solved using gradient image and wavelet transformation to image outline
It is captured with the sensibility of textural characteristics, and the gray level co-occurrence matrixes that different interval distance is utilized simulate the change of human eye viewing distance
The influence of change, described method and system reached with the very high consistency of human visual system, make non-reference picture quality appraisement
Method can simulate subjective visual cognition impression and be evaluated.
Detailed description of the invention
By reference to the following drawings, exemplary embodiments of the present invention can be more fully understood by:
Fig. 1 is a kind of non-reference picture quality appraisement method based on human visual system of the specific embodiment of the invention
Flow chart；
Fig. 2 is a kind of non-reference picture quality appraisement system based on human visual system of the specific embodiment of the invention
Structure chart.
Specific embodiment
Exemplary embodiments of the present invention are introduced referring now to the drawings, however, the present invention can use many different shapes
Formula is implemented, and is not limited to the embodiment described herein, and to provide these embodiments be at large and fully disclose
The present invention, and the scope of the present invention is sufficiently conveyed to person of ordinary skill in the field.Show for what is be illustrated in the accompanying drawings
Term in example property embodiment is not limitation of the invention.In the accompanying drawings, identical cells/elements use identical attached
Icon note.
Unless otherwise indicated, term (including scientific and technical terminology) used herein has person of ordinary skill in the field
It is common to understand meaning.Further it will be understood that with the term that usually used dictionary limits, should be understood as and its
The context of related fields has consistent meaning, and is not construed as Utopian or too formal meaning.
Fig. 1 is a kind of non-reference picture quality appraisement method based on human visual system of the specific embodiment of the invention
Flow chart；The gray level co-occurrence matrixes that different interval distance is utilized in the method simulate the shadow of human eye viewing distance variation
It rings, solves the sensibility capture to image outline and textural characteristics using gradient image and wavelet transformation；Described one kind is based on
The non-reference picture quality appraisement method of human visual system includes:
Step 101, its corresponding characteristic image is calculated according to training image, the characteristic image include training image and
Its corresponding gradient image, horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image；
Training image is denoted as I (x, y), gradient template both horizontally and vertically can be denoted as T respectivelyxAnd Ty；Wherein,
TyFor TxTransposition, Tx=[- 10 1]；
By TxAnd TyConvolution is carried out with I (x, y) respectively, available gradient component both horizontally and vertically is denoted as respectively
GxAnd Gy:
Gx=Tx*I；Gy=Ty*I
Therefore the calculation method for calculating its corresponding gradient image according to distorted image is
Further, the wavelet transformation that I (x, y) is carried out to 1 scale, horizontal, vertical and oblique high fdrequency component is remembered respectively
For HL, LH and HH；Obtain horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image；
Step 102, the gray level co-occurrence matrixes of each characteristic image are calculated；
The image gray levels of the characteristic image are quantified as L grades, and the gray level i and gray level j angle after quantization are θ, and two
Spacing distance is d between a pixel, then the gray level co-occurrence matrixes are expressed as [P (i, j, d, θ)]L*L；
Wherein, 1≤i≤L, 1≤j≤L；I, j and L is natural number；0 °≤θ≤360 °, the value quantity of θ is M；d
Value quantity be it is N number of, d is positive number；The then characteristic image quantity number that the quantity of the gray level co-occurrence matrixes is M*N times；
In the present embodiment, for the directional sensitivity of the prominent vision system of the mankind, θ value quantity is 2, is respectively set
For 0 ° and 90 °.The value number of spacing distance d be 4, respectively 1,2,4 and 8, for simulating the variation of viewing distance.L quilt
Use as default 8.Since image fault understands the luminance information of damaged image simultaneously, also it is extracted the spy of distorted image
Sign.Finally, distorted image, the gradient map of distorted image and 1 scale wavelet transform are extracted under 4 kinds of spacing distances respectively herein
High frequency subgraph (horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image) horizontal and vertical side
To the gray level co-occurrence matrixes of (θ value is set as 0 ° and 90 °), 40 gray level co-occurrence matrixes are obtained.
Step 103, the feature vector of each gray level co-occurrence matrixes is calculated separately；Described eigenvector includes correlation, right
Than property, energy and local entropy；
Step 104, establish feature vector to image quality score mapping relations model；The mapping relations model uses
Support vector regression is established；
Further, the mapping relations model of described eigenvector to image quality score is repeated according to multiple training images
The multiple groups feature vector that step 1 is obtained to 3 is established；
Step 105, the corresponding gray level co-occurrence matrixes feature vector of testing image is brought into the mapping relations model, is obtained
To the mass fraction of testing image；
It in the present embodiment, will be by it in LIVE II, CSIQ and TID2013 number in order to verify the performance of the method for the present invention
According to being tested on library.Choose 3 general performance indicators in image quality evaluation field, i.e., Pearson correlation coefficients (PLCC),
Spearman's correlation coefficient (SROCC) and root-mean-square error (RMSE).Experimental method is provided that divides entire database at random
For 80% training set and 20% test set；The feature vector of training set image is extracted, and by its feature vector and accordingly
Image subjective quality scores input in support vector regression, and training obtains the mapping relations of feature vector and subjective quality scores
Model；Using this regression model, the quality of test image is predicted, and is carried out using prediction score and subjective quality scores
Compare, PLCC, SROCC and RMSE value is calculated；The above process is repeated 1000 times, takes the intermediate value of each performance parameter as this
The final performance of inventive method.
The method of the present invention is tested on LIVE II, CSIQ and the entire database of TID2013 respectively, and obtained performance refers to
Mark is as shown in table 1, it can be seen that RMSE value is smaller, illustrates that the prediction result of the method for the present invention is more accurate.In 3 databases
In upper test result, PLCC and SROCC value is achieved with subjective evaluation result very in 0.93 the method for the present invention described above
Unanimously.Table 2 is the SROCC value that the test in 3 databases on single distortion map image set obtains.SROCC value shown in as can be seen that
0.91 or more, illustrate that the present invention when for single distorted image quality evaluation, equally achieves extraordinary evaluation result.
Test result on 1 LIVE II of table, CSIQ and the entire database of TID2013
Single type of distortion test result on 2 LIVE II of table, CSIQ and TID2013 database
Fig. 2 is a kind of non-reference picture quality appraisement system based on human visual system of the specific embodiment of the invention
Structure chart.The system is based on by technical modellings human visual system, described one kind such as gray level co-occurrence matrixes, wavelet transformations
The non-reference picture quality appraisement system of human visual system includes:
Model generation module 201 and prediction of quality module 202；
The model generation module 201 is used to calculate the gray level co-occurrence matrixes of characteristic image, and according to each gray scale symbiosis
The feature vector of matrix establishes feature vector to the mapping relations model of image quality score；The characteristic image includes training figure
Picture and the corresponding gradient image of training image, horizontal high frequency component image, vertical high frequency component image and oblique high frequency division
Spirogram picture；The model generation module sends the mapping relations model that feature vector establishes feature vector to image quality score
To prediction of quality module；
The prediction of quality module 202 is used for gray level co-occurrence matrixes feature vector and the mapping according to testing image
Relational model obtains the mass fraction of testing image；
Further, the model generation module 201 establishes feature vector to picture quality using support vector regression
The mapping relations model of score；
Further, the training image is one or more；Characteristic image described in each training image is one group corresponding；
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (10)
1. a kind of non-reference picture quality appraisement method based on human visual system, which comprises
Step 1, its corresponding characteristic image is calculated according to training image, the characteristic image includes training image and its correspondence
Gradient image, horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image；
Step 2, the gray level co-occurrence matrixes of each characteristic image are calculated；
Step 3, the feature vector of each gray level co-occurrence matrixes is calculated separately；
Step 4, establish feature vector to image quality score mapping relations model；
Step 5, the corresponding gray level co-occurrence matrixes feature vector of testing image is brought into the mapping relations model, is obtained to be measured
The mass fraction of image.
2. according to the method described in claim 1, it is characterized by: described calculate its corresponding gradient image according to distorted image
Calculation method be
Wherein, G is gradient image, GxFor the gradient component of horizontal direction, GyFor the gradient component of vertical direction；The GxPass through
Gradient template is obtained with distorted image convolution, the GyIt is obtained by the transposition and distorted image convolution of gradient template；The ladder
Spending template is [- 10 1].
3. according to the method described in claim 1, it is characterized by: to the training image carry out 1 scale wavelet transformation, obtain
To horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image.
4. according to the method described in claim 1, it is characterized by: the image gray levels of the characteristic image are quantified as L grades, amount
Gray level i and gray level j angle after change are θ, and spacing distance is d between two pixels, then the gray level co-occurrence matrixes indicate
For [P (i, j, d, θ)]L*L；
Wherein, 1≤i≤L, 1≤j≤L；I, j and L is natural number；0 °≤θ≤360 °, the value quantity of θ is M；D's takes
Be worth quantity be it is N number of, d is positive number；The then characteristic image quantity number that the quantity of the gray level co-occurrence matrixes is M*N times.
5. according to the method described in claim 1, it is characterized by: described eigenvector include correlation, comparative, energy with
And local entropy.
6. according to the method described in claim 1, it is characterized by: establishing feature vector to image using support vector regression
The mapping relations model of mass fraction.
7. according to the method described in claim 1, it is characterized by: described eigenvector to image quality score mapping relations
Model repeats the multiple groups feature vector that step 1 is obtained to 3 according to multiple training images and establishes.
8. a kind of non-reference picture quality appraisement system based on human visual system, the system comprises:
Model generation module and prediction of quality module；
The model generation module is used to calculate the gray level co-occurrence matrixes of characteristic image, and according to the spy of each gray level co-occurrence matrixes
Sign vector establishes feature vector to the mapping relations model of image quality score；The characteristic image includes training image and instruction
Practice the corresponding gradient image of image, horizontal high frequency component image, vertical high frequency component image and oblique high fdrequency component image；Institute
Stating model generation module, that the mapping relations model that feature vector establishes feature vector to image quality score is sent to quality is pre-
Survey module；
The prediction of quality module is used for gray level co-occurrence matrixes feature vector and the mapping relations mould according to testing image
Type obtains the mass fraction of testing image.
9. system according to claim 8, it is characterised in that: the model generation module is built using support vector regression
Mapping relations model of the vertical feature vector to image quality score.
10. according to the method described in claim 8, it is characterized by: the training image is one or more；Each training figure
The characteristic image as described in one group corresponding.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711286394.1A CN109919894A (en) | 2017-12-07 | 2017-12-07 | A kind of non-reference picture quality appraisement method and system based on human visual system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711286394.1A CN109919894A (en) | 2017-12-07 | 2017-12-07 | A kind of non-reference picture quality appraisement method and system based on human visual system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109919894A true CN109919894A (en) | 2019-06-21 |
Family
ID=66956628
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711286394.1A Pending CN109919894A (en) | 2017-12-07 | 2017-12-07 | A kind of non-reference picture quality appraisement method and system based on human visual system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919894A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110532511A (en) * | 2019-07-26 | 2019-12-03 | 四川师范大学 | It is a kind of based on gray scale-order co-occurrence matrix rotary inertia powder concentration measurement method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090116713A1 (en) * | 2007-10-18 | 2009-05-07 | Michelle Xiao-Hong Yan | Method and system for human vision model guided medical image quality assessment |
| CN106408561A (en) * | 2016-09-10 | 2017-02-15 | 天津大学 | Texture feature-based image quality evaluating method without reference |
| CN106780449A (en) * | 2016-12-06 | 2017-05-31 | 嘉兴学院 | A kind of non-reference picture quality appraisement method based on textural characteristics |
| CN107145855A (en) * | 2017-04-28 | 2017-09-08 | 努比亚技术有限公司 | One kind is without reference mass blurred picture Forecasting Methodology, terminal and storage medium |
| CN107343196A (en) * | 2017-07-18 | 2017-11-10 | 天津大学 | One kind mixing distortion non-reference picture quality appraisement method |
-
2017
- 2017-12-07 CN CN201711286394.1A patent/CN109919894A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090116713A1 (en) * | 2007-10-18 | 2009-05-07 | Michelle Xiao-Hong Yan | Method and system for human vision model guided medical image quality assessment |
| CN106408561A (en) * | 2016-09-10 | 2017-02-15 | 天津大学 | Texture feature-based image quality evaluating method without reference |
| CN106780449A (en) * | 2016-12-06 | 2017-05-31 | 嘉兴学院 | A kind of non-reference picture quality appraisement method based on textural characteristics |
| CN107145855A (en) * | 2017-04-28 | 2017-09-08 | 努比亚技术有限公司 | One kind is without reference mass blurred picture Forecasting Methodology, terminal and storage medium |
| 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 (3)
| Title |
|---|
| RUOZHANG YANG ET.AL: "No-reference image quality assessment based on deep learning method", 《2017 IEEE 3RD INFORMATION TECHNOLOGY AND MECHATRONICS ENGINEERING CONFERENCE (ITOEC)》 * |
| 侯春萍: "基于高阶相位一致性的混合失真图像质量评价", 《激光与光电子学进展》 * |
| 叶正汕: "基于部分参考和无参考的水下图像传输质量的评测方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110532511A (en) * | 2019-07-26 | 2019-12-03 | 四川师范大学 | It is a kind of based on gray scale-order co-occurrence matrix rotary inertia powder concentration measurement method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108428227B (en) | No-reference image quality evaluation method based on full convolution neural network | |
| CN103996192B (en) | Non-reference image quality evaluation method based on high-quality natural image statistical magnitude model | |
| CN107105223B (en) | A kind of tone mapping method for objectively evaluating image quality based on global characteristics | |
| CN105049851B (en) | General non-reference picture quality appraisement method based on Color perception | |
| CN101976444B (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
| CN106373124B (en) | Industrial products surface defect visible detection method based on gray level co-occurrence matrixes and RANSAC | |
| CN107396095B (en) | A kind of no reference three-dimensional image quality evaluation method | |
| CN106709958A (en) | Gray scale gradient and color histogram-based image quality evaluation method | |
| CN105574901B (en) | A kind of general non-reference picture quality appraisement method based on local contrast pattern | |
| CN104021545A (en) | Full-reference color image quality evaluation method based on visual saliency | |
| CN107507166A (en) | It is a kind of based on support vector regression without refer to screen image quality measure method | |
| CN106408561A (en) | Texture feature-based image quality evaluating method without reference | |
| CN102075786A (en) | Method for objectively evaluating image quality | |
| CN106454350A (en) | Non-reference evaluation method for infrared image | |
| CN109978854A (en) | A kind of screen content image quality measure method based on edge and structure feature | |
| CN102547368A (en) | Objective evaluation method for quality of stereo images | |
| CN109544504A (en) | Screen picture quality evaluating method based on rarefaction representation | |
| CN107343196A (en) | One kind mixing distortion non-reference picture quality appraisement method | |
| CN108550146A (en) | A kind of image quality evaluating method based on ROI | |
| CN106023214B (en) | Image quality evaluating method and system based on central fovea view gradient-structure similitude | |
| CN112184672A (en) | No-reference image quality evaluation method and system | |
| CN110996096A (en) | Tone mapping image quality evaluation method based on structural similarity difference | |
| CN104346809A (en) | Image quality evaluation method for image quality dataset adopting high dynamic range | |
| CN107018410B (en) | A kind of non-reference picture quality appraisement method based on pre- attention mechanism and spatial dependence | |
| CN103996188B (en) | A kind of full-reference image quality evaluating method based on Gabor weighted features |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190621 |
|
| RJ01 | Rejection of invention patent application after publication |