CN109685757A - A kind of non-reference picture quality appraisement method and system based on grey scale difference statistics - Google Patents
A kind of non-reference picture quality appraisement method and system based on grey scale difference statistics Download PDFInfo
- Publication number
- CN109685757A CN109685757A CN201710973078.5A CN201710973078A CN109685757A CN 109685757 A CN109685757 A CN 109685757A CN 201710973078 A CN201710973078 A CN 201710973078A CN 109685757 A CN109685757 A CN 109685757A
- Authority
- CN
- China
- Prior art keywords
- image
- gray scale
- scale difference
- training data
- distorted 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
- G06T2207/30—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)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of non-reference picture quality appraisement methods based on grey scale difference statistics, include: that n times low-pass filtering and down-sampling processing are carried out continuously to each distorted image that the training data is concentrated according to preset times N, obtains the 2nd to N+1 scale image of each distorted image that the training data is concentrated respectively;Calculate training data concentrate each distorted image and the corresponding 2nd to N+1 scale image local binary patterns map;The gray scale difference value of the local binary patterns map is calculated, and the gray scale difference value is counted, obtains gray scale difference probability-distribution function;Based on gray scale difference probability-distribution function, statistic is constructed, obtains feature vector;The subjective quality scores of described eigenvector and each distorted image are trained using support vector regression SVM, determine the mapping relations model of feature vector and subjective quality scores;The mass fraction of test image is evaluated using the mapping relations model, obtains evaluation result.
Description
Technical field
The present invention relates to picture research fields, and more particularly, to it is a kind of based on grey scale difference statistics without reference
Image quality evaluating method and system.
Background technique
The development of Internet technology and universal, enriches the daily of people, facilitates interpersonal communication, special
It is not a large amount of transmission of the multimedia content such as image and video, brings great visual enjoyment to us.But in image
During acquisition, transimission and storage etc., image is usually introduced into different type and different degrees of distortion, leads to picture quality
Decline, affects the viewing effect of image.In order to improve the performance of image processing system and image delivering system, need to image
Quality effectively judged, therefore, it is particularly significant to establish rationally effective image quality evaluating method.
Image quality evaluation algorithm can be divided into subjective picture quality evaluation method and Objective image quality evaluation method.It is main
It sees image quality evaluating method to require subject direct viewing image and evaluate the quality of image, this method time-consuming consumption
Power, therefore be not suitable for applying in actual image procossing and image delivering system.Objective image quality evaluation method is by building
Vertical mathematical model, calculates the quality of image, the Real-Time Evaluation of great amount of images quality may be implemented.According to available ginseng
Examine image information number, image quality evaluating method can be divided into following three kinds: full reference, half reference and non-reference picture matter
Measure evaluation method.In engineer application, original image can not be generally obtained, therefore, non-reference picture quality appraisement is image matter
The emphasis of amount evaluation area research.
Summary of the invention
The present invention provides a kind of non-reference picture quality appraisement methods and system based on grey scale difference statistics, to solve
To the determination problem of image quality evaluation.
To solve the above-mentioned problems, according to an aspect of the invention, there is provided a kind of nothing based on grey scale difference statistics
Reference image quality appraisement method, which comprises
N times low-pass filtering is carried out continuously under to each distorted image that the training data is concentrated according to preset times N
Sampling processing obtains the 2nd to N+1 scale image of each distorted image that the training data is concentrated respectively;
Calculate training data concentrate each distorted image and the corresponding 2nd to N+1 scale image local binary patterns
Map;
Calculate each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image local binary
The gray scale difference value of mode map, and the gray scale difference value is counted, obtain gray scale difference probability-distribution function;
Based on gray scale difference probability-distribution function, statistic is constructed, obtains each distorted image that the training data is concentrated
With the corresponding 2nd to N+1 scale image feature vector;
Each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image feature vector and
The subjective quality scores of each distorted image are trained using support vector regression SVM, determine feature vector and subjective matter
Measure the mapping relations model of score;
It is carried out using mass fraction of the mapping relations model of described eigenvector and subjective quality scores to test image
Evaluation obtains evaluation result.
Preferably, wherein the preset times are 4 times.
Preferably, wherein the radius that is local and knowing mode is 1, neighborhood point number is 8.
Preferably, wherein each distorted image that the training data is concentrated and the corresponding 2nd that calculates is to N+1 scale
The gray scale difference score value of the local binary patterns map of image, and the gray scale fractionation value is counted, obtain gray scale difference probability
Distribution function, comprising:
Remember that distorted image is I (x, y), corresponding local binary patterns map is Ilbp, (x, y) is IlbpIn image a bit, point (x,
Y) corresponding gray value is g (x, y), is divided between (x, y)The gray value of point beThen
The gray scale difference value of two pixels are as follows:
Each distorted image and the corresponding 2nd that training data is concentrated are obtained to N+1 scale image using the above method
The gray scale difference value of all pixels point;
All possible values of the gray scale difference value are counted, are divided between acquisitionGray scale difference is
Gray scale difference probability-distribution function.
Preferably, wherein the interval
Preferably, wherein described eigenvector includes: contrast, angular second moment, entropy and average value,
The calculation formula of the contrast are as follows:
The calculation formula of the angular second moment are as follows:
The calculation formula of the entropy are as follows:
The calculation formula of the average value are as follows:
Wherein, CON is contrast, and ASM is angular second moment, and ENT is entropy, and MEAN is average value, and i is gray scale difference value,
It is divided into 1 for, gray scale difference value is the gray scale difference probability-distribution function of i.
Another invention according to the present invention, provides a kind of non-reference picture quality appraisement based on grey scale difference statistics
System, the system comprises: multiple scale image acquisition unit, binary pattern map acquiring unit, gray scale difference probability distribution letter
Number acquiring unit, feature vector acquiring unit, mapping relations model determination unit and evaluation unit,
The multiple scale image acquisition unit, each mistake for being concentrated according to preset times N to the training data
True image is carried out continuously n times low-pass filtering and down-sampling processing, obtains each distorted image that the training data is concentrated respectively
The the 2nd to N+1 scale image;
The binary pattern map acquiring unit, for calculating each distorted image and corresponding the of training data concentration
2 to N+1 scale image local binary patterns map;
The gray scale difference probability-distribution function acquiring unit, each distorted image concentrated for calculating the training data
With the gray scale difference value of the corresponding 2nd to N+1 scale image local binary patterns map, and unite to the gray scale difference value
Meter obtains gray scale difference probability-distribution function;
Described eigenvector acquiring unit constructs statistic, obtains the instruction for being based on gray scale difference probability-distribution function
Practice data set in each distorted image and the corresponding 2nd to N+1 scale image feature vector;
The mapping relations model determination unit, each distorted image for concentrating the training data and corresponding
2nd to N+1 scale image feature vector and each distorted image subjective quality scores using support vector regression SVM into
Row training, determines the mapping relations model of feature vector and subjective quality scores;
The evaluation unit, for the mapping relations model using described eigenvector and subjective quality scores to test chart
The mass fraction of picture is evaluated, and evaluation result is obtained.
Preferably, wherein the preset times are 4 times.
Preferably, wherein the radius that is local and knowing mode is 1, and neighborhood point number is 8.
Preferably, wherein the gray scale difference probability-distribution function acquiring unit, calculates each of described training data concentration
Distorted image and the corresponding 2nd to N+1 scale image local binary patterns map gray scale difference score value, and the gray scale is torn open
Score value is counted, and gray scale difference probability-distribution function is obtained, comprising:
Remember that distorted image is I (x, y), corresponding local binary patterns map is Ilbp, (x, y) is IlbpIn image a bit, point
(x, y) corresponding gray value is g (x, y), is divided between (x, y)The gray value of point be
The then gray scale difference value of two pixels are as follows:
Each distorted image and the corresponding 2nd that training data is concentrated are obtained to N+1 scale image using the above method
The gray scale difference value of all pixels point;
All possible values of the gray scale difference value are counted, are divided between acquisitionGray scale difference isGray scale difference probability-distribution function.
Preferably, wherein the interval
Preferably, wherein described eigenvector includes: contrast, angular second moment, entropy and average value,
The calculation formula of the contrast are as follows:
The calculation formula of the angular second moment are as follows:
The calculation formula of the entropy are as follows:
The calculation formula of the average value are as follows:
Wherein, CON is contrast, and ASM is angular second moment, and ENT is entropy, and MEAN is average value, and i is gray scale difference value,
It is divided into 1 for, gray scale difference value is the gray scale difference probability-distribution function of i.
The present invention provides a kind of non-reference picture quality appraisement methods and system based on grey scale difference statistics, pass through meter
Calculate the local binary patterns characteristic pattern of distorted image;It calculates the grey scale difference of local binary patterns image and counts acquisition gray scale difference
Probability-distribution function;The subjective quality scores of feature vector and each distorted image are utilized branch by the feature vector for obtaining image
It holds vector regression SVM to be trained, determines the mapping relations model of feature vector and subjective quality scores, and reflect described in utilization
Relational model is penetrated to evaluate picture quality.The image quality evaluation result of the method for the present invention and human visual system have very
High consistency effectively has rated the quality of image, has to the performance improvement of image processing system and image delivering system
Very big realistic meaning.
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 the non-reference picture quality appraisement method 100 based on grey scale difference statistics according to embodiment of the present invention
Schematic diagram;And
Fig. 2 is the non-reference picture quality appraisement system 200 based on grey scale difference statistics according to embodiment of the present invention
Structural schematic diagram.
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 the non-reference picture quality appraisement method 100 based on grey scale difference statistics according to embodiment of the present invention
Schematic diagram.As shown in Figure 1, the non-reference picture quality appraisement method based on grey scale difference statistics of embodiment of the present invention
100, each distorted image that the training data is concentrated is carried out continuously at n times low-pass filtering and down-sampling according to preset times
Reason obtains the 2nd to N+1 scale image of each distorted image that the training data is concentrated respectively;Training data is calculated to concentrate
Each distorted image and the corresponding 2nd to N+1 scale image local binary patterns map;Calculate local binary patterns map
Gray scale difference value, and the gray scale difference value is counted, obtains gray scale difference probability-distribution function;Based on gray scale difference probability distribution
Function constructs statistic, obtains each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image
Feature vector;Each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image feature vector and
The subjective quality scores of each distorted image are trained using support vector regression SVM, determine feature vector and subjective matter
Measure the mapping relations model of score;Using the mapping relations model of described eigenvector and subjective quality scores to test image
Mass fraction is evaluated, and evaluation result is obtained.Image texture is a kind of important characteristics of image, and the variation of image texture
It is closely related with image fault type, image fault degree.Therefore, the line that embodiment according to the present invention passes through extraction image
Feature is managed, image quality evaluation model is established using support vector regression SVM, realizes non-reference picture quality appraisement.It is described
Method 100 is since step 101 place, in each distorted image that step 101 concentrates the training data according to preset times N
Be carried out continuously n times low-pass filtering and down-sampling processing, obtain respectively the 2nd of each distorted image that the training data is concentrated to
N+1 scale image.Preferably, wherein the preset times are 4 times.Preferably, wherein the radius that is local and knowing mode is
1, neighborhood point number is 8.
Preferably, each distorted image that training data is concentrated and the corresponding 2nd is calculated to N+1 scalogram in step 102
The local binary patterns map of picture.
Preferably, each distorted image that the training data is concentrated and the corresponding 2nd is calculated to N+1 ruler in step 103
The gray scale difference value of the local binary patterns map of image is spent, and the gray scale difference value is counted, obtains gray scale difference probability point
Cloth function.
Preferably, wherein each distorted image that the training data is concentrated and the corresponding 2nd that calculates is to N+1 scale
The gray scale difference score value of the local binary patterns map of image, and the gray scale fractionation value is counted, obtain gray scale difference probability
Distribution function, comprising:
Remember that distorted image is I (x, y), corresponding local binary patterns map is Ilbp, (x, y) is IlbpIn image a bit, point
(x, y) corresponding gray value is g (x, y), is divided between (x, y)The gray value of point be
The then gray scale difference value of two pixels are as follows:
Each distorted image and the corresponding 2nd that training data is concentrated are obtained to N+1 scale image using the above method
The gray scale difference value of all pixels point;
All possible values of the gray scale difference value are counted, are divided between acquisitionGray scale difference is
Gray scale difference probability-distribution function.Preferably, wherein the interval
Preferably, it is based on gray scale difference probability-distribution function in step 104, constructs statistic, obtains the training dataset
In each distorted image and the corresponding 2nd to N+1 scale image feature vector.Preferably, wherein described eigenvector packet
It includes: contrast, angular second moment, entropy and average value,
The calculation formula of the contrast are as follows:
The calculation formula of the angular second moment are as follows:
The calculation formula of the entropy are as follows:
The calculation formula of the average value are as follows:
Wherein, CON is contrast, and ASM is angular second moment, and ENT is entropy, and MEAN is average value, and i is gray scale difference value,
It is divided into 1 for, gray scale difference value is the gray scale difference probability-distribution function of i.
In embodiments of the present invention, be based on gray scale difference probability-distribution function, respectively construct contrast, angular second moment,
4 statistics of entropy and average value, the calculation method of 4 statistics and the characteristics of image of description are specific as follows:
Contrast describes the depth of image definition and image texture rill.Image is more clear, and rill is deeper, then compares
The value of degree is bigger.Conversely, fuzzy image contrast value is small.Contrast is denoted as CON, calculation method such as following formula:
Angular second moment reflects the uniformity coefficient of the intensity profile of image, and the intensity profile of image is more uniform, angular second moment
It is smaller.Angular second moment is denoted as ASM, calculation formula is such as shown in (3):
Entropy has measured the information content of image, when entropy is larger, shows that image distribution is uneven.Entropy is denoted as ENT, it is specific to calculate
Method is as follows:
Average value reflects the overall distribution situation of gray value of image, and the luminance area of image is more, and average value is bigger.It will
Average value is denoted as MEAN, and formula is as follows:
Four low-pass filtering and down-sampling are carried out continuously to each distorted image in training set, are respectively obtained in training set
Each distorted image the 2nd, 3,4 and 5 scale image, four feature extractions are respectively completed to the image on 5 scales
Journey, the final feature vector to each distorted image are 20 dimensions.
Preferably, each distorted image for concentrating the training data in step 105 and the corresponding 2nd is to N+1 scale
The subjective quality scores of the feature vector of image and each distorted image are trained using support vector regression SVM, are determined
The mapping relations model of feature vector and subjective quality scores.
Preferably, utilize the mapping relations model of described eigenvector and subjective quality scores to test chart in step 106
The mass fraction of picture is evaluated, and evaluation result is obtained.
In order to verify the non-reference picture quality appraisement method based on grey scale difference statistics of embodiments of the present invention
Performance tests it on II database of LIVE.Using 3 universal performance index test sheets in image quality evaluation field
The performance of inventive method, i.e. Pearson correlation coefficients (PLCC), Spearman's correlation coefficient (SROCC) and root-mean-square error
(RMSE).Wherein, the monotonicity that SROCC value measures objective prediction score and subjective assessment score indicates evaluation closer to 1
As a result better.PLCC value characterizes the correlation of objective prediction score and subjective assessment score, and value indicates prediction score closer to 1
It is higher with subjective quality scores correlation.Root-mean-square error is the absolute mistake between objective prediction score and subjective assessment score
Difference, root-mean-square error show that objective prediction is more accurate closer to 0.
Experimental method of the invention is provided that takes 80% image as training set at random in II database of LIVE,
Remaining image is test set;The feature vector for extracting training set and test set image, by the subjective quality scores of training set image and
It is trained study in image feature vector input support vector regression, obtains the mapping of feature vector and subjective quality scores
Relational model;The feature vector of test image is inputted into this regression model, obtains the forecast quality score of test image;It calculates pre-
PLCC, SROCC and RMSE value of mass metering score and subjective quality scores;The above process is repeated 1000 times, each performance is taken to join
Final performance of several intermediate values as the non-reference picture quality appraisement method based on grey scale difference statistics of present embodiment.
3 performance indicators tested on II database of LIVE are as shown in table 1, it can be seen that RMSE value is smaller, says
The image quality evaluation result of bright the method for the present invention is more accurate.Test obtains on single distortion data collection and entire database
PLCC and SROCC value illustrates that image quality evaluation result and subjective evaluation result of the invention are more consistent 0.89 or more.
PLCC, SROCC and RMSE intermediate value on 1 LIVE of table, II database in 1000 iteration tests
The experimental results showed that image quality evaluation model evaluation result of the invention is accurate, it is consistent with human visual system
Property it is high, can effectively evaluate the quality of image, have to the performance improvement of image processing system and image delivering system very big
Realistic meaning.
Fig. 2 is the non-reference picture quality appraisement system 200 based on grey scale difference statistics according to embodiment of the present invention
Structural schematic diagram.As shown in Fig. 2, the non-reference picture quality appraisement system 200 based on grey scale difference statistics includes: more
Subdimension image acquisition unit 201, binary pattern map acquiring unit 202, gray scale difference probability-distribution function acquiring unit 203,
Feature vector acquiring unit 204, mapping relations model determination unit 205 and evaluation unit 206.
Preferably, the multiple scale image acquisition unit 201 is used for according to preset times N to the training dataset
In each distorted image be carried out continuously n times low-pass filtering and down-sampling processing, obtain respectively the training data concentrate it is every
2nd to N+1 scale image of a distorted image.Preferably, wherein the preset times are 4 times.
Preferably, wherein the radius that is local and knowing mode is 1, and neighborhood point number is 8.
Preferably, the binary pattern map acquiring unit 202, for calculating each distorted image of training data concentration
With the corresponding 2nd to N+1 scale image local binary patterns map.
Preferably, the gray scale difference probability-distribution function acquiring unit 203, for calculating the every of the training data concentration
A distorted image and the corresponding 2nd to N+1 scale image local binary patterns map gray scale difference value, and to the gray scale difference
Value is counted, and gray scale difference probability-distribution function is obtained.Preferably, wherein the gray scale difference probability-distribution function acquiring unit,
Calculate each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image local binary patterns map
Gray scale difference score value, and the gray scale fractionation value is counted, obtains gray scale difference probability-distribution function, comprising:
Remember that distorted image is I (x, y), corresponding local binary patterns map is Ilbp, (x, y) is IlbpIn image a bit, point
(x, y) corresponding gray value is g (x, y), is divided between (x, y)The gray value of point be
The then gray scale difference value of two pixels are as follows:
Each distorted image and the corresponding 2nd that training data is concentrated are obtained to N+1 scale image using the above method
The gray scale difference value of all pixels point;
All possible values of the gray scale difference value are counted, are divided between acquisitionGray scale difference isGray scale difference probability-distribution function.
Preferably, wherein the interval
Preferably, described eigenvector acquiring unit 204, for constructing statistic based on gray scale difference probability-distribution function,
Obtain each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image feature vector.Preferably,
Wherein described eigenvector includes: contrast, angular second moment, entropy and average value,
The calculation formula of the contrast are as follows:
The calculation formula of the angular second moment are as follows:
The calculation formula of the entropy are as follows:
The calculation formula of the average value are as follows:
Wherein, CON is contrast, and ASM is angular second moment, and ENT is entropy, and MEAN is average value, and i is gray scale difference value,
It is divided into 1 for, gray scale difference value is the gray scale difference probability-distribution function of i.
Preferably, the mapping relations model determination unit 205, each distortion map for concentrating the training data
Picture and the corresponding 2nd to N+1 scale image feature vector and each distorted image subjective quality scores utilize supporting vector
Regression machine SVM is trained, and determines the mapping relations model of feature vector and subjective quality scores.
Preferably, the evaluation unit 206, for the mapping relations mould using described eigenvector and subjective quality scores
Type evaluates the mass fraction of test image, obtains evaluation result.
The non-reference picture quality appraisement system 200 and of the invention based on grey scale difference statistics of the embodiment of the present invention
The non-reference picture quality appraisement method 100 based on grey scale difference statistics of another embodiment is corresponding, and details are not described herein.
The present invention is described by reference to a small amount of embodiment.However, it is known in those skilled in the art, as
Defined by subsidiary Patent right requirement, in addition to the present invention other embodiments disclosed above equally fall in it is of the invention
In range.
Normally, all terms used in the claims are all solved according to them in the common meaning of technical field
It releases, unless in addition clearly being defined wherein.All references " one/described/be somebody's turn to do [device, component etc.] " are all opened ground
At least one example being construed in described device, component etc., unless otherwise expressly specified.Any method disclosed herein
Step need not all be run with disclosed accurate sequence, unless explicitly stated otherwise.
Claims (12)
1. a kind of non-reference picture quality appraisement method based on grey scale difference statistics, which is characterized in that the described method includes:
N times low-pass filtering and down-sampling are carried out continuously to each distorted image that the training data is concentrated according to preset times N
Processing obtains the 2nd to N+1 scale image of each distorted image that the training data is concentrated respectively;
Calculate training data concentrate each distorted image and the corresponding 2nd to N+1 scale image local binary patterns map;
Calculate each distorted image that the training data is concentrated and the corresponding 2nd to N+1 scale image local binary patterns
The gray scale difference value of map, and the gray scale difference value is counted, obtain gray scale difference probability-distribution function;
Based on gray scale difference probability-distribution function, statistic is constructed, obtains each distorted image that the training data is concentrated and right
The 2nd answered to N+1 scale image feature vector;
Each distorted image that the training data is concentrated and the corresponding 2nd is to the feature vector of N+1 scale image and each
The subjective quality scores of distorted image are trained using support vector regression SVM, determine feature vector and subjective quality point
Several mapping relations models;
The mass fraction of test image is evaluated using the mapping relations model of described eigenvector and subjective quality scores,
Obtain evaluation result.
2. the method according to claim 1, wherein the preset times are 4 times.
3. the method according to claim 1, wherein it is described part and know mode radius be 1, neighborhood point number
It is 8.
4. the method according to claim 1, wherein each distortion map for calculating the training data and concentrating
Picture and the corresponding 2nd to N+1 scale image local binary patterns map gray scale difference score value, and to the gray scale fractionation be worth into
Row statistics, obtains gray scale difference probability-distribution function, comprising:
Remember that distorted image is I (x, y), corresponding local binary patterns map is Ilbp, (x, y) is IlbpIn image a bit, point (x,
Y) corresponding gray value is g (x, y), is divided between (x, y)The gray value of point beThen
The gray scale difference value of two pixels are as follows:
The each distorted image and the corresponding 2nd owning to N+1 scale image that training data is concentrated are obtained using the above method
The gray scale difference value of pixel;
All possible values of the gray scale difference value are counted, are divided between acquisitionGray scale difference is
Gray scale difference probability-distribution function.
5. according to the method described in claim 4, it is characterized in that, the interval
6. according to the method described in claim 5, it is characterized in that, described eigenvector includes: contrast, angular second moment, entropy
And average value,
The calculation formula of the contrast are as follows:
The calculation formula of the angular second moment are as follows:
The calculation formula of the entropy are as follows:
The calculation formula of the average value are as follows:
Wherein, CON is contrast, and ASM is angular second moment, and ENT is entropy, and MEAN is average value, and i is gray scale difference value,For
It is divided into 1, gray scale difference value is the gray scale difference probability-distribution function of i.
7. a kind of non-reference picture quality appraisement system based on grey scale difference statistics, which is characterized in that the system comprises: it is more
Subdimension image acquisition unit, binary pattern map acquiring unit, gray scale difference probability-distribution function acquiring unit, feature vector obtain
Unit, mapping relations model determination unit and evaluation unit are taken,
The multiple scale image acquisition unit, each distortion map for being concentrated according to preset times N to the training data
As being carried out continuously n times low-pass filtering and down-sampling processing, the 2nd of each distorted image that the training data is concentrated is obtained respectively
To N+1 scale image;
The binary pattern map acquiring unit, for calculating each distorted image and the corresponding 2nd to N of training data concentration
The local binary patterns map of+1 scale image;
The gray scale difference probability-distribution function acquiring unit, for calculating each distorted image that the training data is concentrated and right
The gray scale difference value of the local binary patterns map for the 2nd to the N+1 scale image answered, and the gray scale difference value is counted, it obtains
Take gray scale difference probability-distribution function;
Described eigenvector acquiring unit constructs statistic, obtains the trained number for being based on gray scale difference probability-distribution function
According to concentration each distorted image and the corresponding 2nd to N+1 scale image feature vector;
The mapping relations model determination unit, each distorted image for concentrating the training data and the corresponding 2nd
The subjective quality scores of feature vector and each distorted image to N+1 scale image are carried out using support vector regression SVM
Training, determines the mapping relations model of feature vector and subjective quality scores;
The evaluation unit, for the mapping relations model using described eigenvector and subjective quality scores to test image
Mass fraction is evaluated, and evaluation result is obtained.
8. the method according to the description of claim 7 is characterized in that the preset times are 4 times.
9. the method according to the description of claim 7 is characterized in that it is described part and know mode radius be 1, neighborhood point number
It is 8.
10. the method according to the description of claim 7 is characterized in that the gray scale difference probability-distribution function acquiring unit, calculates
The ash of each distorted image and the corresponding 2nd local binary patterns map to N+1 scale image that the training data is concentrated
Difference value is spent, and the gray scale fractionation value is counted, obtains gray scale difference probability-distribution function, comprising:
Remember that distorted image is I (x, y), corresponding local binary patterns map is Ilbp, (x, y) is IlbpIn image a bit, point (x, y)
Corresponding gray value is g (x, y), is divided between (x, y)The gray value of point beThen two
The gray scale difference value of a pixel are as follows:
The each distorted image and the corresponding 2nd owning to N+1 scale image that training data is concentrated are obtained using the above method
The gray scale difference value of pixel;
All possible values of the gray scale difference value are counted, are divided between acquisitionGray scale difference is
Gray scale difference probability-distribution function.
11. according to the method described in claim 10, it is characterized in that, the interval
12. according to the method for claim 11, which is characterized in that described eigenvector include: contrast, angular second moment,
Entropy and average value,
The calculation formula of the contrast are as follows:
The calculation formula of the angular second moment are as follows:
The calculation formula of the entropy are as follows:
The calculation formula of the average value are as follows:
Wherein, CON is contrast, and ASM is angular second moment, and ENT is entropy, and MEAN is average value, and i is gray scale difference value,For
It is divided into 1, gray scale difference value is the gray scale difference probability-distribution function of i.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710973078.5A CN109685757B (en) | 2017-10-18 | 2017-10-18 | Non-reference image quality evaluation method and system based on gray difference statistics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710973078.5A CN109685757B (en) | 2017-10-18 | 2017-10-18 | Non-reference image quality evaluation method and system based on gray difference statistics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109685757A true CN109685757A (en) | 2019-04-26 |
CN109685757B CN109685757B (en) | 2021-04-02 |
Family
ID=66183807
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710973078.5A Active CN109685757B (en) | 2017-10-18 | 2017-10-18 | Non-reference image quality evaluation method and system based on gray difference statistics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109685757B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111597284A (en) * | 2020-04-29 | 2020-08-28 | 武汉子雨科技有限公司 | Method for rapidly extracting surveying and mapping geographic information data |
CN111652854A (en) * | 2020-05-13 | 2020-09-11 | 中山大学 | No-reference image quality evaluation method based on image high-frequency information |
Citations (4)
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 |
CN104902267A (en) * | 2015-06-08 | 2015-09-09 | 浙江科技学院 | No-reference image quality evaluation method based on gradient information |
CN106408561A (en) * | 2016-09-10 | 2017-02-15 | 天津大学 | Texture feature-based image quality evaluating method without reference |
CN106600597A (en) * | 2016-12-22 | 2017-04-26 | 华中科技大学 | Non-reference color image quality evaluation method based on local binary pattern |
-
2017
- 2017-10-18 CN CN201710973078.5A patent/CN109685757B/en active Active
Patent Citations (4)
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 |
CN104902267A (en) * | 2015-06-08 | 2015-09-09 | 浙江科技学院 | No-reference image quality evaluation method based on gradient information |
CN106408561A (en) * | 2016-09-10 | 2017-02-15 | 天津大学 | Texture feature-based image quality evaluating method without reference |
CN106600597A (en) * | 2016-12-22 | 2017-04-26 | 华中科技大学 | Non-reference color image quality evaluation method based on local binary pattern |
Non-Patent Citations (2)
Title |
---|
PEDRO GARCIA FREITAS 等: "No-Reference Image Quality Assessment Using Texture Information Banks", 《2016 5TH BRAZILIAN CONFERENCE ON INTELLIGENT SYSTEMS》 * |
刘辉 等: "基于灰度差分统计的火焰图像纹理特征提取", 《控制工程》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111597284A (en) * | 2020-04-29 | 2020-08-28 | 武汉子雨科技有限公司 | Method for rapidly extracting surveying and mapping geographic information data |
CN111652854A (en) * | 2020-05-13 | 2020-09-11 | 中山大学 | No-reference image quality evaluation method based on image high-frequency information |
Also Published As
Publication number | Publication date |
---|---|
CN109685757B (en) | 2021-04-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108428227B (en) | No-reference image quality evaluation method based on full convolution neural network | |
CN110046673B (en) | No-reference tone mapping image quality evaluation method based on multi-feature fusion | |
CN103996192B (en) | Non-reference image quality evaluation method based on high-quality natural image statistical magnitude model | |
CN105407349B (en) | No-reference objective three-dimensional image quality evaluation method based on binocular visual perception | |
CN108090902A (en) | A kind of non-reference picture assessment method for encoding quality based on multiple dimensioned generation confrontation network | |
CN110378232B (en) | Improved test room examinee position rapid detection method of SSD dual-network | |
CN108021938A (en) | A kind of Cold-strip Steel Surface defect online detection method and detecting system | |
CN101562675B (en) | No-reference image quality evaluation method based on Contourlet transform | |
CN110728656A (en) | Meta-learning-based no-reference image quality data processing method and intelligent terminal | |
CN106709958A (en) | Gray scale gradient and color histogram-based image quality evaluation method | |
CN102075786A (en) | Method for objectively evaluating image quality | |
CN103971340A (en) | High-bit-width digital image dynamic range compression and detail enhancement method | |
CN101976444A (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
CN108053396A (en) | A kind of more distorted image quality without with reference to evaluation method | |
CN104036493A (en) | No-reference image quality evaluation method based on multifractal spectrum | |
CN107343196A (en) | One kind mixing distortion non-reference picture quality appraisement method | |
CN104346809A (en) | Image quality evaluation method for image quality dataset adopting high dynamic range | |
CN102722888A (en) | Stereoscopic image objective quality evaluation method based on physiological and psychological stereoscopic vision | |
CN109272044A (en) | A kind of image similarity determines method, apparatus, equipment and storage medium | |
CN106934770A (en) | A kind of method and apparatus for evaluating haze image defog effect | |
CN109685757A (en) | A kind of non-reference picture quality appraisement method and system based on grey scale difference statistics | |
CN104574424B (en) | Based on the nothing reference image blur evaluation method of multiresolution DCT edge gradient statistics | |
CN115908142A (en) | Contact net tiny part damage testing method based on visual recognition | |
CN114596287A (en) | Image quality determination method, device, equipment and storage medium | |
CN106022362A (en) | Reference-free image quality objective evaluation method for JPEG2000 compression distortion |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |