CN107481236A - A kind of quality evaluating method of screen picture - Google Patents
A kind of quality evaluating method of screen picture Download PDFInfo
- Publication number
- CN107481236A CN107481236A CN201710748668.8A CN201710748668A CN107481236A CN 107481236 A CN107481236 A CN 107481236A CN 201710748668 A CN201710748668 A CN 201710748668A CN 107481236 A CN107481236 A CN 107481236A
- Authority
- CN
- China
- Prior art keywords
- formula
- screen picture
- feature
- screen
- quality
- 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
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/60—Extraction of image or video features relating to illumination properties, e.g. using a reflectance or lighting model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a kind of quality evaluating method of screen picture,Objective quality scores are exported come evaluation image quality by the visual signature and mapping that extract screen picture,Local contrast normalization operation is passed through to the screen picture of distortion,Obtain luminance graph,And therefrom obtain the brightness statistics feature of screen picture,And gradient map is calculated based on luminance graph,Gradient map obtains Texture Statistical Feature through local binary patterns,By in brightness statistics feature and Texture Statistical Feature input support vector regression SVR,Obtain the mapping model between visual signature and subjective quality scores,The visual signature of any screen picture is inputted in the mapping model,Export as objective quality scores,Index using objective quality scores as evaluation image quality,Realize the screen picture quality evaluation algorithm of supervision,Gained evaluation result more meets the perception of human visual system.
Description
Technical field
The invention belongs to multimedia technology field, and in particular to a kind of quality evaluating method of screen picture.
Background technology
Screen picture is widely used in the various multimedia application in our daily lifes, such as computer and intelligence
Information sharing system between phone, cloud computing system, the visual quality of screen picture have very big to the viewing experience of user
Influence.Image be human perception and Machinery model identification important information source, adequacy and standard of its quality to acquired information
True property plays conclusive effect, but generates image during the collection in image, compression, processing, transmission and display etc.
Degradation problems, how effective image quality evaluation machine is established in fields such as transmission of video, character recognition, security monitoring, medical science
Fixture is of great importance.
Traditional visual quality appraisal procedure includes Y-PSNR and mean square error, and the subjective sensation of itself and people differ
Cause, because it does not consider the feature of human visual system, however, the main body of image quality evaluation is people, researcher regards to human eye
Feel that the research of characteristic is related to physiology, psychology etc., so far to it still without fully understanding, the psychological characteristic particularly to vision
Also it is difficult to find out quantitative description method, so also lacking a set of comprehensive, unified image quality evaluation system at present.In order to
Overcome these shortcomings, propose advanced visual quality evaluation method in the prior art, including structural similarity, feature phase
Like property and gradient similarity deviation etc..
The method of image quality evaluation has:(1) subjective evaluation method, by contrived experiment, by observer to picture quality
Evaluated;(2) method for objectively evaluating, picture quality is evaluated using algorithm.Wherein subjective evaluation method is by observing
Person provides Quality estimation to testing image according to prespecified opinion scale or the experience of oneself, but time-consuming, complicated, can also
Influenceed by subjective factors such as observer's specialty background, psychology and motivations, and not can be incorporated into other algorithms and use.It is objective
Evaluation method is convenient, fast, easily realizes and can be incorporated into application system, but has deviation with the subjective feeling of people, conventional
Image quality evaluation algorithm is objective evaluation algorithm, and its target is to obtain the objective evaluation consistent with subjective evaluation result
Value.
Objective evaluation algorithm is divided into three classes according to its degree of dependence to reference picture:(1) full reference picture quality is commented
Estimate, it is necessary to which the full detail of original image is as reference;(2) half reference picture quality evaluations, it is only necessary to and on reference picture
Partial information compares;(3) non-reference picture quality evaluation, it is not necessary to any reference information.
Non-reference picture quality appraisement advantage is not need the information of original image, and just can carry out quality to distorted image comments
Valency, transinformation is considerably reduced, its algorithm difficult point is:Characteristics of image is difficult to define and extracted, and human eye, which perceives, to be difficult to
Modelling represents.It can be divided into the algorithm for type of distortion and the algorithm based on machine learning without algorithm is referred to.For distortion
The algorithm of type, specific algorithm are turn of peak value and energy from high frequency to low frequency in some characteristic frequencies according to distorted image
In-migration weighs its distortion level;Algorithm based on machine learning, specific algorithm is will be special in bottom by the method for machine learning
Image similar in the quality with identical rule is classified as one kind in sign, meanwhile, machine learning need not analyze the type of distortion with
And feature how is extracted to weigh the degree of distortion, but the mark of obtained " feature " directly as image quality evaluation will be learnt
Standard, this kind of algorithm is first by image block, using the pixel value in every piece of region as original feature vector, then using different
Machine learning strategy therefrom extracts the strong new feature vector of classification capacity, finally utilizes new feature vector with reference to subjective assessment value
Classified or returned.It is evaluation result is not influenceed as far as possible by picture material without the difficult point with reference to algorithm, calculates
The foundation that method carries out quality evaluation is some characteristic statistics extracted from image, and these statistics are easily by picture material
Influence.
The content of the invention
In order to solve deficiency of the prior art, the present invention provides a kind of quality evaluating method of screen picture.
The technical scheme is that:A kind of quality evaluating method of screen picture, it is characterised in that from screen picture
Quality assessment database, extract the visual signature of distorted screen image and corresponding subjective quality scores, wherein visual signature bag
Brightness statistics feature and Texture Statistical Feature are included, obtains mapping mould during visual signature and subjective quality scores input vector are returned
Type, the visual signature of arbitrary image, which is inputted in mapping model, can obtain objective quality scores, be used as with objective quality scores and commented
The index of valency picture quality, obtain mapping model the step of include:
A. luminance graph is obtained, and therefrom extract screen picture by local contrast normalization operation to the screen picture of distortion
Brightness statistics feature;
B. the luminance graph based on gained, gradient map, i.e. first-order derivative characteristic figure is calculated using gradient operator;
C. according to gained gradient map, second dervative characteristic pattern is calculated to obtain by local binary patterns, and from second dervative characteristic pattern
The Texture Statistical Feature of middle extraction screen picture;
D. in gained visual signature and corresponding subjective quality scores input support vector regression SVR, visual signature and master are obtained
Mapping model between appearance quality fraction.
Preferably, it is characterised in that described screen picture quality assessment database is database SIQAD.
Preferably, the visual signature of the extraction screen picture uses and 3 different sizes of image is extracted with visual signature, and 3
Individual different sizes are full size, 1/2 size after a down-sampling is handled and 1/4 size after the processing of secondary down-sampling.
Preferably, horizontal directions are tieed up in brightness statistics feature of the brightness statistics feature including 10 dimension screen pictures, 10
Brightness statistics feature, the brightness statistics feature of 10 dimension vertical direction, the brightness statistics feature and 10 dimensions in 10 dimension leading diagonal directions
The brightness statistics feature in minor diagonal direction.
Preferably, vertical direction are tieed up in Texture Statistical Feature of the Texture Statistical Feature including 10 dimension horizontal directions, 10
The Texture Statistical Feature of Texture Statistical Feature, the Texture Statistical Feature in 10 dimension leading diagonal directions and 10 dimension minor diagonal directions.
Further, it is described to extract comprising the concrete steps that for brightness statistics feature:
A. contrast part normalization operation is carried out to distorted screen image P and obtains luminance graph, as shown in formula (1):
Wherein, coordinate is the pixel of (i, j) in P (i.j) expressions screen picture,Represent the screen after normalized
The brightness value at (i, j) place in image, C are constant, C=6.5025,WithThe equal of pixel regional area pixel value is represented respectively
Value and variance,WithCalculate as shown in formula (2), formula (3):
Wherein,, M=N=3;
B. luminance graph is passed throughBrightness statistics feature is calculated, 10 dimensional vectors are expressed as in the form of histogram, histogram calculation is such as
Shown in formula (4), formula (5):
Wherein,The absolute value of pixel point value in luminance graph is represented,Represent the value model of each post of histogram
Enclose, I and J represent the width and height of image respectively, it can thus be concluded that brightness statistics characteristic vector 10 dimensional feature vectors, for representing monochrome information;
C. brightness statistics feature, including horizontal direction are represented using the histogram feature of four different directions of screen picture
(H), vertical direction (V), leading diagonal direction (D1) and minor diagonal direction (D2), such as formula (6), formula (7), formula (8), formula (9)
It is shown:
Wherein,、、、Respectively horizontal direction, vertical direction, leading diagonal direction, minor diagonal
The brightness statistics feature in direction,For the brightness value at (i, j) place in screen picture,。
Further, it is described to extract comprising the concrete steps that for Texture Statistical Feature:
A. in order to obtain the directional information of screen picture, definition respectively represents horizontal, vertical, leading diagonal, minor diagonal four
The gradient operator of different directions、、、, gradient map is calculated by convolution algorithm according to luminance graph, as formula (10),
Shown in formula (11), formula (12), formula (13):
Wherein,Screen intensity figure is represented,、、、Gradient map, as one corresponding to four different directions are represented respectively
Order derivative characteristic pattern;
B. the uniform local binary patterns of invariable rotary are respectively adopted to the gradient map of four direction and calculate to obtain second dervative feature
Figure, as shown in formula (14), formula (15), formula (16):
Wherein,For second dervative characteristic pattern, K is the number of neighborhood point at pixel, and R is between pixel and neighborhood point
Distance, K=8, R=1,The brightness value of pixel is represented,The brightness value of K symmetric neighborhood point is represented respectively,For the local binary patterns 0-1 of pixel transition times,The brightness value of regional area center pixel is represented, with histogram
Form by formula (4), formula (5) extracted respectively from second dervative information four direction 10 dimension Texture Statistical Features.
Further, in the input support vector regression SVR, by the visual signature extracted and corresponding subjective quality
Fraction inputs SVR, using RBF as kernel function, can obtain the mapping mould that visual signature maps to subjective quality scores
Type, step are as follows:
A. the screen picture of S width distortions is included in database, the subjective quality scores of every width screen picture are obtained by subjective experiment
Obtain, the subjective quality scores of image are represented with DMOS, the subjective quality scores of note m width screen pictures are, wherein 1
≤ m≤S, the visual feature vector of every width distorted screen image is extracted, usedRepresent the visual signatures of m width screen pictures to
Amount, whereinDimension be 270;
B. useThe data base set of S width distorted screen images is represented,, forIn m-th of feature to
Amount, using RBFAs kernel function, its support vector regression SVR mapping model g (), such as formula
(17), formula(18)It is shown:
Wherein, 1≤j≤S,For weight vector, T is the transposition of vector, and b is constant,For the objective quality point of output
Number, exp are exponential function using natural constant e as the truth of a matter, "" to calculate Euclidean distance symbol,For the nuclear parameter of function,;
C. with mapping model pairIn the visual signatures of all screen pictures be trained, training objective is makesValue and its
It is correspondingIt is worth closest, gained optimized parameter isWith, then final gained mapping model such as formula(19)It is shown:
Wherein,For the objective quality scores of output.
Compared with prior art, the beneficial effects of the invention are as follows:
(1) full size to screen picture, 1/2 size and 1/4 size extraction visual signature, make extracted visual signature more accord with
Close human visual perception characteristic;
(2) while to screen picture and horizontal direction, vertical direction, main diagonally opposed, secondary diagonally opposed image zooming-out brightness
Statistical nature, to horizontal direction, vertical direction, main diagonally opposed, secondary diagonally opposed image zooming-out Texture Statistical Feature, make institute
The visual signature precision of extraction is higher;
(3) 10 dimensional vectors are expressed as by the form of histogram to brightness statistics feature and Texture Statistical Feature, more conventional is logical
The fitting of definition generalized Gaussian distribution is crossed, its algorithm is simple, and extraction brightness statistics feature efficiency is higher.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention;
Fig. 2 is the gradient operator of horizontal direction;
Fig. 3 is the gradient operator of vertical direction;
Diagonally opposed gradient operator based on Fig. 4;
Fig. 5 is secondary diagonally opposed gradient operator。
Embodiment
The present invention is described in further detail below in conjunction with accompanying drawing.
As shown in figure 1, a kind of quality evaluating method of screen picture, it is characterised in that from screen picture quality evaluation
Database SIQAD, the visual signature of distorted screen image and corresponding subjective quality scores are extracted, wherein visual signature includes bright
Statistical nature and Texture Statistical Feature are spent, mapping model is obtained during visual signature and subjective quality scores input vector are returned,
Objective quality scores can be obtained in the visual signature input mapping model of arbitrary image, are used as evaluation figure with objective quality scores
As the index of quality, the step of obtaining mapping model, includes:
A. luminance graph is obtained, and therefrom extract screen picture by local contrast normalization operation to the screen picture of distortion
Brightness statistics feature;
B. the luminance graph based on gained, gradient map, i.e. first-order derivative characteristic figure is calculated using gradient operator;
C. according to gained gradient map, second dervative characteristic pattern is calculated to obtain by local binary patterns, and from second dervative characteristic pattern
The Texture Statistical Feature of middle extraction screen picture;
D. in gained visual signature and corresponding subjective quality scores input support vector regression SVR, visual signature and master are obtained
Mapping model between appearance quality fraction.
The visual signature of the extraction screen picture uses extracts visual signature, 3 different chis to 3 different sizes of image
Very little 1/2 size for full size, after a down-sampling is handled and 1/4 size after the processing of secondary down-sampling.
The extraction brightness statistics feature comprises the concrete steps that:
A. contrast part normalization operation is carried out to distorted screen image P and obtains luminance graph, as shown in formula (1):
Wherein, coordinate is the pixel of (i, j) in P (i.j) expressions screen picture,Represent the screen after normalized
The brightness value at (i, j) place in image, C are constant, C=6.5025,WithThe equal of pixel regional area pixel value is represented respectively
Value and variance,WithCalculate as shown in formula (2), formula (3):
Wherein,, M=N=3;
B. luminance graph is passed throughBrightness statistics feature is calculated, 10 dimensional vectors are expressed as in the form of histogram, histogram calculation is such as
Shown in formula (4), formula (5):
Wherein,The absolute value of pixel point value in luminance graph is represented,Represent the value model of each post of histogram
Enclose, I and J represent the width and height of image respectively, it can thus be concluded that brightness statistics characteristic vector 10 dimensional feature vectors, for representing monochrome information;
C. brightness statistics feature, including horizontal direction are represented using the histogram feature of four different directions of screen picture
(H), vertical direction (V), leading diagonal direction (D1) and minor diagonal direction (D2), such as formula (6), formula (7), formula (8), formula (9)
It is shown:
Wherein,、、、Respectively horizontal direction, vertical direction, leading diagonal direction, minor diagonal side
To brightness statistics feature,For the brightness value at (i, j) place in screen picture,。
The extraction Texture Statistical Feature comprises the concrete steps that:
A. in order to obtain the directional information of screen picture, definition respectively represents horizontal, vertical, leading diagonal, minor diagonal four
The gradient operator of different directions、、、, gradient map is calculated by convolution algorithm according to luminance graph, as formula (10),
Shown in formula (11), formula (12), formula (13):
Wherein,Screen intensity figure is represented,、、、Gradient map, as one corresponding to four different directions are represented respectively
Order derivative characteristic pattern;
B. the uniform local binary patterns of invariable rotary are respectively adopted to the gradient map of four direction and calculate to obtain second dervative feature
Figure, as shown in formula (14), formula (15), formula (16):
Wherein,For second dervative characteristic pattern, K is the number of neighborhood point at pixel, and R is between pixel and neighborhood point
Distance, K=8, R=1,The brightness value of pixel is represented,The brightness value of K symmetric neighborhood point is represented respectively,For the local binary patterns 0-1 of pixel transition times,The brightness value of regional area center pixel is represented, with histogram
Form by formula (4), formula (5) extracted respectively from second dervative information four direction 10 dimension Texture Statistical Features.
In the input support vector regression SVR, the visual signature extracted and corresponding subjective quality scores are inputted
SVR, using RBF as kernel function, the mapping model that visual signature maps to subjective quality scores, step can be obtained
It is as follows:
A. the screen picture of S width distortions is included in database, the subjective quality scores of every width screen picture are obtained by subjective experiment
Obtain, the subjective quality scores of image are represented with DMOS, the subjective quality scores of note m width screen pictures are, wherein 1
≤ m≤S, the visual feature vector of every width distorted screen image is extracted, usedRepresent the visual signatures of m width screen pictures to
Amount, whereinDimension be 270;
B. useThe data base set of S width distorted screen images is represented,, forIn m-th of feature to
Amount, using RBFAs kernel function, its support vector regression SVR mapping model g (), such as formula
(17), formula(18)It is shown:
Wherein, 1≤j≤S,For weight vector, T is the transposition of vector, and b is constant,For the objective quality point of output
Number, exp are exponential function using natural constant e as the truth of a matter, "" to calculate Euclidean distance symbol,For the nuclear parameter of function,;
C. with mapping model pairIn the visual signatures of all screen pictures be trained, training objective is makesValue and its
It is correspondingIt is worth closest, gained optimized parameter isWith, then final gained mapping model such as formula(19)It is shown:
Wherein,For the objective quality scores of output.
In order to further illustrate the feasibility of the inventive method and validity, using Pearson's linearly dependent coefficient PLCC,
Spearman rank correlation coefficient SROCC and root-mean-square error RMSE refers to as the test of screen picture quality evaluating method performance
Mark, corresponding subjective quality is each provided with using the screen picture in screen picture quality database SIQAD, SIQAD database
Fraction, the evaluation to screen picture is performed on SIQAD databases and tests the performance of this method.
Database is divided into training set and test set, wherein training intensive data accounts for the 80% of total Database, test set accounts for
20%, by training the visual signature and corresponding subjective quality scores that intensive data extracted, obtain visual signature and map to master
The mapping model of appearance quality fraction, the performance of the inventive method is examined by test set.In order to remove the influence of randomness, to this
Operation performs 1000 times, takes performance of the intermediate result as the inventive method.
During using test set to detect the mapping model, in the visual signature input mapping model that test set is extracted,
The objective quality scores of test set are obtained, in order to calculate the performance of various evaluation methods under same yardstick, to objective quality point
Number is changed, and using the subjective quality scores of test set as ordinate, objective quality scores are abscissa, and fitting obtains subjective matter
The functional relation between fraction and objective quality scores is measured, such as formula(20)It is shown:
Wherein, β1、β2、β3、β4And β5It is fitting constant, the objective quality scores of test set is substituted into formula again(20)In, can
Objective quality scores are carried out to be converted to the objective quality scores after changing.
It is calculated by the objective quality scores after the subjective quality scores, objective quality scores and conversion of test set
PLCC, SROCC and RMSE value, such as formula(21), formula(22)With formula (23) Suo Shi:
Wherein,WithThe prediction subjective quality scores and subjective quality scores of the i-th width image are represented respectively,WithRespectively
The average value corresponding to it is represented,WithThe sequence of subjective quality scores and objective quality scores in the sequence is represented respectively
Number.What PLCC coefficient correlations reflected is the accuracy of evaluating objective quality algorithm prediction, and accuracy is higher, its order of magnitude
Closer to 1, otherwise closer to 0;What SROCC reflected is the monotonicity of evaluating objective quality algorithm prediction, and its value is closer
It is higher in 1 explanation monotonicity, it is lower closer to 0 explanation monotonicity;Root-mean-square error value is smaller, and accuracy is higher, on the contrary, value is got over
Greatly, accuracy is lower.
Following examples do not limit the present invention to preferably explain the present invention.
Embodiment 1
Evaluation method of the invention and existing full reference image quality appraisement model are performed respectively in database SIQAD to screen
Curtain image evaluated, full reference image quality appraisement model include PSNR models, SSIM models, VIF models, MAD models,
GMSD models, SPQA models, SQI models, GSS models and EMSQA models, are tested evaluation result, gained PLCC values and
SROCC values are as shown in table 1, it can be seen that the present invention commonly uses full reference mass evaluation model more in the prior art has the excellent of protrusion
Gesture, quality evaluation effect are more preferable.
Embodiment 2
Evaluation method of the invention and existing non-reference picture quality appraisement model are performed respectively in database SIQAD to screen
Curtain image evaluated, non-reference picture quality appraisement model include NIQE models, ILNIQE models, BRISQUE models,
GMLOG models and BQMS models, are tested evaluation result, and gained PLCC values and SROCC values are as shown in table 2, it can be seen that
The present invention commonly uses reference-free quality evaluation model more in the prior art equally has prominent advantage.
Reference examples 1
Following steps, and test quality evaluation result are performed in database SIQAD, as shown in table 3:
A. luminance graph is obtained, and therefrom extract screen picture by local contrast normalization operation to the screen picture of distortion
Brightness statistics feature;
B. extracted from database SIQAD distorted image brightness statistics feature and distorted screen image corresponding to subjective quality point
Number, will in brightness statistics feature and corresponding subjective quality scores input support vector regression SVR, obtain brightness statistics feature with
Mapping model between subjective quality scores, the brightness statistics feature of any screen picture are inputted in the mapping model, are exported and are
Objective quality scores, the index using objective quality scores as evaluation image quality.
Table 1
Packet | PLCC values | SROCC values | RMSE value |
PSNR models | 0.5869 | 0.5608 | 11.5859 |
SSIM models | 0.5912 | 0.5836 | 11.545 |
VIF models | 0.8206 | 0.8069 | 8.1795 |
MAD models | 0.6191 | 0.6067 | 11.2409 |
GMSD models | 0.7259 | 0.7305 | 9.4684 |
SPQA models | 0.8584 | 0.8416 | 7.3421 |
SQI models | 0.8644 | 0.8548 | 7.1782 |
GSS models | 0.8461 | 0.8359 | 7.631 |
EMSQA models | 0.8648 | 0.8504 | 7.186 |
The present invention | 0.8442 | 0.8202 | 7.5957 |
Table 2
Packet | PLCC values | SROCC values | RMSE value |
NIQE models | 0.3749 | 0.3568 | 13.152 |
ILNIQE models | 0.3854 | 0.3212 | 13.2085 |
BRISQUE models | 0.8113 | 0.7749 | 8.2565 |
GMLOG models | 0.7608 | 0.7035 | 9.253 |
BQMS models | 0.8115 | 0.8005 | 9.3042 |
The present invention | 0.8442 | 0.8202 | 7.5957 |
Reference examples 2
Following steps, and test quality evaluation result are performed in database SIQAD, as shown in table 3:
A. luminance graph is obtained by local contrast normalization operation to the screen picture of distortion, the luminance graph based on gained,
Gradient map is calculated using gradient operator;
B. according to gained gradient map, second dervative characteristic pattern is calculated to obtain by local binary patterns, and from second dervative characteristic pattern
The Texture Statistical Feature of middle extraction screen picture;
C. subjective matter corresponding to the Texture Statistical Feature and distorted screen image of distorted screen image is extracted from database SIQAD
Fraction is measured, Texture Statistical Feature and corresponding subjective quality scores are inputted in support vector regression SVR, obtains texture statistics spy
Mapping model between sign and subjective quality scores, the Texture Statistical Feature of any screen picture is inputted in the mapping model, defeated
Go out for objective quality scores, the index using objective quality scores as evaluation image quality.
Table 3
Packet | Reference examples 1 | Reference examples 2 | The present invention |
PLCC values | 0.8333 | 0.7781 | 0.8442 |
SROCC values | 0.7936 | 0.7245 | 0.8202 |
RMSE value | 7.8341 | 8.9273 | 7.5957 |
The test value of reference examples 1, reference examples 2 and the present invention is contrasted, reference examples 1 are only with brightness statistics feature to screen picture
Quality evaluation is carried out, reference examples 2 carry out quality evaluation only with Texture Statistical Feature, and the present invention is compared with reference examples 1 and reference examples 2
Evaluation result be respectively provided with obvious advantage, i.e., in image quality evaluation, brightness statistics feature and Texture Statistical Feature are to commenting
The quality no less important of valency image, while progress image quality evaluation can improve the accuracy of evaluation result both use.
The desirable embodiment according to the present invention is enlightenment above, and by above-mentioned description, related personnel completely can be with
Without departing from the scope of the technological thought of the present invention', various changes and amendments are carried out.The technical scope of this invention
The content being not limited on specification, it is necessary to determine the technical scope according to the scope of the claims.
Claims (8)
1. a kind of quality evaluating method of screen picture, it is characterised in that from screen picture quality assessment database, from data
The visual signature of distorted screen image and corresponding subjective quality scores are extracted in storehouse, it is special to include brightness statistics for wherein visual signature
Seek peace Texture Statistical Feature, will visual signature and subjective quality scores input vector return in obtain mapping model, arbitrary image
Visual signature input mapping model in can obtain objective quality scores, be used as evaluation image quality with objective quality scores
Index, obtain mapping model the step of include:
A. luminance graph is obtained, and therefrom extract screen picture by local contrast normalization operation to the screen picture of distortion
Brightness statistics feature;
B. the luminance graph based on gained, gradient map, i.e. first-order derivative characteristic figure is calculated using gradient operator;
C. according to gained gradient map, second dervative characteristic pattern is calculated to obtain by local binary patterns, and from second dervative characteristic pattern
The Texture Statistical Feature of middle extraction screen picture;
D. in gained visual signature and corresponding subjective quality scores input support vector regression SVR, visual signature and master are obtained
Mapping model between appearance quality fraction.
A kind of 2. quality evaluating method of screen picture according to claim 1, it is characterised in that described screen picture
Quality assessment database is database SIQAD.
A kind of 3. quality evaluating method of screen picture according to claim 1, it is characterised in that the extraction screen map
The visual signature of picture uses extracts visual signature to 3 different sizes of image, and 3 different sizes are full size, through once adopting down
1/2 size after sample processing and 1/4 size after the processing of secondary down-sampling.
4. the quality evaluating method of a kind of screen picture according to claim 1, it is characterised in that the brightness statistics are special
Sign include 10 dimension screen pictures brightness statistics feature, 10 dimension horizontal directions brightness statistics feature, 10 tie up vertical direction it is bright
Spend the brightness statistics feature of statistical nature, the brightness statistics feature in 10 dimension leading diagonal directions and 10 dimension minor diagonal directions.
5. the quality evaluating method of a kind of screen picture according to claim 1, it is characterised in that the texture statistics is special
Sign includes the Texture Statistical Feature of 10 dimension horizontal directions, the Texture Statistical Feature of 10 dimension vertical direction, 10 dimension leading diagonal directions
Texture Statistical Feature and 10 dimension minor diagonal directions Texture Statistical Feature.
6. the quality evaluating method of a kind of screen picture according to claim 1 or 4, it is characterised in that the extraction is bright
Degree statistical nature comprises the concrete steps that:
A. contrast part normalization operation is carried out to distorted screen image P and obtains luminance graph, as shown in formula (1):
Wherein, coordinate is the pixel of (i, j) in P (i.j) expressions screen picture,Represent the screen after normalized
The brightness value at (i, j) place in image, C are constant, C=6.5025,WithThe equal of pixel regional area pixel value is represented respectively
Value and variance,WithCalculate as shown in formula (2), formula (3):
Wherein,, M=N=3;
B. luminance graph is passed throughBrightness statistics feature is calculated, 10 dimensional vectors are expressed as in the form of histogram, histogram calculation is such as
Shown in formula (4), formula (5): 1
Wherein,The absolute value of pixel point value in luminance graph is represented,The span of each post of histogram is represented,
I and J represents the width and height of image respectively, it can thus be concluded that brightness statistics characteristic vector 10 dimensional feature vectors, for representing monochrome information;
C. brightness statistics feature, including horizontal direction are represented using the histogram feature of four different directions of screen picture
(H), vertical direction (V), leading diagonal direction (D1) and minor diagonal direction (D2), such as formula (6), formula (7), formula (8), formula (9)
It is shown:
Wherein,、、、Respectively horizontal direction, vertical direction, leading diagonal direction, secondary diagonal
The brightness statistics feature in line direction,For the brightness value at (i, j) place in screen picture,。
A kind of 7. quality evaluating method of screen picture according to claim 1 or 5, it is characterised in that the extraction line
Reason statistical nature comprises the concrete steps that:
A. in order to obtain the directional information of screen picture, definition respectively represents horizontal, vertical, leading diagonal, minor diagonal four
The gradient operator of different directions、、、, gradient map is calculated by convolution algorithm according to luminance graph, as formula (10),
Shown in formula (11), formula (12), formula (13):
Wherein,Screen intensity figure is represented,、、、Gradient map, as single order corresponding to four different directions are represented respectively
Derivative characteristic pattern;
B. the uniform local binary patterns of invariable rotary are respectively adopted to the gradient map of four direction and calculate to obtain second dervative feature
Figure, as shown in formula (14), formula (15), formula (16):
2
Wherein,For second dervative characteristic pattern, K is the number of neighborhood point at pixel, and R is between pixel and neighborhood point
Distance, K=8, R=1,The brightness value of pixel is represented,The brightness value of K symmetric neighborhood point is represented respectively,For the local binary patterns 0-1 of pixel transition times,The brightness value of regional area center pixel is represented, with histogram
Form by formula (4), formula (5) extracted respectively from second dervative information four direction 10 dimension Texture Statistical Features.
A kind of 8. quality evaluating method of screen picture according to claim 1, it is characterised in that it is described input support to
Amount returns SVR, and the visual signature extracted and corresponding subjective quality scores are inputted into SVR, core is used as using RBF
Function, the mapping model that visual signature maps to subjective quality scores can be obtained, is comprised the following steps that:
A. the screen picture of S width distortions is included in database, the subjective quality scores of every width screen picture are obtained by subjective experiment
Obtain, the subjective quality scores of image are represented with DMOS, the subjective quality scores of note m width screen pictures are, wherein 1
≤ m≤S, the visual feature vector of every width distorted screen image is extracted, usedRepresent the visual signatures of m width screen pictures to
Amount, whereinDimension be 270;
B. useThe data base set of S width distorted screen images is represented,, forIn m-th of feature to
Amount, using RBFAs kernel function, its support vector regression SVR mapping model g (), such as formula
(17), formula(18)It is shown:
Wherein, 1≤j≤S,For weight vector, T is the transposition of vector, and b is constant,For the objective quality point of output
Number, exp are exponential function using natural constant e as the truth of a matter, "" to calculate Euclidean distance symbol,For the nuclear parameter of function,;
C. with mapping model pairIn the visual signatures of all screen pictures be trained, training objective is makesValue and its
It is correspondingIt is worth closest, gained optimized parameter isWith, then final gained mapping model such as formula(19)It is shown:
Wherein,For the objective quality scores of output. 3
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710748668.8A CN107481236A (en) | 2017-08-28 | 2017-08-28 | A kind of quality evaluating method of screen picture |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710748668.8A CN107481236A (en) | 2017-08-28 | 2017-08-28 | A kind of quality evaluating method of screen picture |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107481236A true CN107481236A (en) | 2017-12-15 |
Family
ID=60602769
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710748668.8A Pending CN107481236A (en) | 2017-08-28 | 2017-08-28 | A kind of quality evaluating method of screen picture |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107481236A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109210855A (en) * | 2018-11-29 | 2019-01-15 | 宁波泽锦电器科技有限公司 | Intelligent freon-free refrigerator |
CN109344860A (en) * | 2018-08-19 | 2019-02-15 | 天津大学 | A kind of non-reference picture quality appraisement method based on LBP |
CN109544504A (en) * | 2018-10-16 | 2019-03-29 | 天津大学 | Screen picture quality evaluating method based on rarefaction representation |
CN109978834A (en) * | 2019-03-05 | 2019-07-05 | 方玉明 | A kind of screen picture quality evaluating method based on color and textural characteristics |
CN109978854A (en) * | 2019-03-25 | 2019-07-05 | 福州大学 | A kind of screen content image quality measure method based on edge and structure feature |
CN110796651A (en) * | 2019-10-29 | 2020-02-14 | 杭州阜博科技有限公司 | Image quality prediction method and device, electronic device and storage medium |
CN111047618A (en) * | 2019-12-25 | 2020-04-21 | 福州大学 | Multi-scale-based non-reference screen content image quality evaluation method |
CN111127387A (en) * | 2019-07-11 | 2020-05-08 | 宁夏大学 | Method for evaluating quality of non-reference image |
CN111275681A (en) * | 2020-01-19 | 2020-06-12 | 浙江大华技术股份有限公司 | Picture quality determination method and device, storage medium and electronic device |
CN111553969A (en) * | 2019-12-31 | 2020-08-18 | 深圳积木易搭科技技术有限公司 | Texture mapping method, medium, terminal and device based on gradient domain |
CN114066857A (en) * | 2021-11-18 | 2022-02-18 | 烟台艾睿光电科技有限公司 | Infrared image quality evaluation method and device, electronic equipment and readable storage medium |
WO2024055357A1 (en) * | 2022-09-15 | 2024-03-21 | 深圳Tcl新技术有限公司 | Television definition adjustment method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102750695A (en) * | 2012-06-04 | 2012-10-24 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
CN103200421A (en) * | 2013-04-07 | 2013-07-10 | 北京理工大学 | No-reference image quality evaluation method based on Curvelet transformation and phase coincidence |
CN105894522A (en) * | 2016-04-28 | 2016-08-24 | 宁波大学 | Multi-distortion stereo image quality objective evaluation method |
CN106548472A (en) * | 2016-11-03 | 2017-03-29 | 天津大学 | Non-reference picture quality appraisement method based on Walsh Hadamard transform |
-
2017
- 2017-08-28 CN CN201710748668.8A patent/CN107481236A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102750695A (en) * | 2012-06-04 | 2012-10-24 | 清华大学 | Machine learning-based stereoscopic image quality objective assessment method |
CN103200421A (en) * | 2013-04-07 | 2013-07-10 | 北京理工大学 | No-reference image quality evaluation method based on Curvelet transformation and phase coincidence |
CN105894522A (en) * | 2016-04-28 | 2016-08-24 | 宁波大学 | Multi-distortion stereo image quality objective evaluation method |
CN106548472A (en) * | 2016-11-03 | 2017-03-29 | 天津大学 | Non-reference picture quality appraisement method based on Walsh Hadamard transform |
Non-Patent Citations (2)
Title |
---|
CHANG C C: "A library for support vector machines", 《ACM TRANS.INTELLIGENT SYMP》 * |
YUMING FANG: "No reference quality assessment for screen content images", 《 2017 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA & EXPO WORKSHOPS (ICMEW)》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109344860A (en) * | 2018-08-19 | 2019-02-15 | 天津大学 | A kind of non-reference picture quality appraisement method based on LBP |
CN109544504A (en) * | 2018-10-16 | 2019-03-29 | 天津大学 | Screen picture quality evaluating method based on rarefaction representation |
CN109210855A (en) * | 2018-11-29 | 2019-01-15 | 宁波泽锦电器科技有限公司 | Intelligent freon-free refrigerator |
CN109978834A (en) * | 2019-03-05 | 2019-07-05 | 方玉明 | A kind of screen picture quality evaluating method based on color and textural characteristics |
CN109978854B (en) * | 2019-03-25 | 2022-07-08 | 福州大学 | Screen content image quality evaluation method based on edge and structural features |
CN109978854A (en) * | 2019-03-25 | 2019-07-05 | 福州大学 | A kind of screen content image quality measure method based on edge and structure feature |
CN111127387B (en) * | 2019-07-11 | 2024-02-09 | 宁夏大学 | Quality evaluation method for reference-free image |
CN111127387A (en) * | 2019-07-11 | 2020-05-08 | 宁夏大学 | Method for evaluating quality of non-reference image |
CN110796651A (en) * | 2019-10-29 | 2020-02-14 | 杭州阜博科技有限公司 | Image quality prediction method and device, electronic device and storage medium |
CN111047618B (en) * | 2019-12-25 | 2023-01-03 | 福州大学 | Multi-scale-based non-reference screen content image quality evaluation method |
CN111047618A (en) * | 2019-12-25 | 2020-04-21 | 福州大学 | Multi-scale-based non-reference screen content image quality evaluation method |
CN111553969A (en) * | 2019-12-31 | 2020-08-18 | 深圳积木易搭科技技术有限公司 | Texture mapping method, medium, terminal and device based on gradient domain |
CN111553969B (en) * | 2019-12-31 | 2023-09-05 | 深圳积木易搭科技技术有限公司 | Texture mapping method, medium, terminal and device based on gradient domain |
CN111275681A (en) * | 2020-01-19 | 2020-06-12 | 浙江大华技术股份有限公司 | Picture quality determination method and device, storage medium and electronic device |
CN111275681B (en) * | 2020-01-19 | 2023-09-01 | 浙江大华技术股份有限公司 | Picture quality determining method and device, storage medium and electronic device |
CN114066857A (en) * | 2021-11-18 | 2022-02-18 | 烟台艾睿光电科技有限公司 | Infrared image quality evaluation method and device, electronic equipment and readable storage medium |
WO2024055357A1 (en) * | 2022-09-15 | 2024-03-21 | 深圳Tcl新技术有限公司 | Television definition adjustment method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107481236A (en) | A kind of quality evaluating method of screen picture | |
Kundu et al. | No-reference quality assessment of tone-mapped HDR pictures | |
CN103996192B (en) | Non-reference image quality evaluation method based on high-quality natural image statistical magnitude model | |
CN105744256B (en) | Based on the significant objective evaluation method for quality of stereo images of collection of illustrative plates vision | |
CN109325550B (en) | No-reference image quality evaluation method based on image entropy | |
CN105049851B (en) | General non-reference picture quality appraisement method based on Color perception | |
CN107396095B (en) | A kind of no reference three-dimensional image quality evaluation method | |
CN109255358B (en) | 3D image quality evaluation method based on visual saliency and depth map | |
Yue et al. | Blind stereoscopic 3D image quality assessment via analysis of naturalness, structure, and binocular asymmetry | |
CN107105223B (en) | A kind of tone mapping method for objectively evaluating image quality based on global characteristics | |
CN101976444B (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
CN108805825B (en) | Method for evaluating quality of repositioning image | |
CN107146220B (en) | A kind of universal non-reference picture quality appraisement method | |
CN103325120A (en) | Rapid self-adaption binocular vision stereo matching method capable of supporting weight | |
CN107507166A (en) | It is a kind of based on support vector regression without refer to screen image quality measure method | |
Wang et al. | A new blind image quality framework based on natural color statistic | |
CN112184672A (en) | No-reference image quality evaluation method and system | |
CN104318545A (en) | Foggy weather polarization image quality evaluation method | |
CN102722888A (en) | Stereoscopic image objective quality evaluation method based on physiological and psychological stereoscopic vision | |
CN109754390A (en) | A kind of non-reference picture quality appraisement method based on mixing visual signature | |
CN111641822B (en) | Method for evaluating quality of repositioning stereo image | |
CN105894507B (en) | Image quality evaluating method based on amount of image information natural scene statistical nature | |
Wang et al. | Evaluation on synthesized face sketches | |
Chen et al. | Blind stereo image quality assessment based on binocular visual characteristics and depth perception | |
Wang et al. | Distortion recognition for image quality assessment with convolutional neural network |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171215 |