CN108230325A - The compound degraded image quality evaluating method and system decomposed based on cartoon texture - Google Patents
The compound degraded image quality evaluating method and system decomposed based on cartoon texture Download PDFInfo
- Publication number
- CN108230325A CN108230325A CN201810119012.4A CN201810119012A CN108230325A CN 108230325 A CN108230325 A CN 108230325A CN 201810119012 A CN201810119012 A CN 201810119012A CN 108230325 A CN108230325 A CN 108230325A
- Authority
- CN
- China
- Prior art keywords
- image
- texture
- noise
- evaluation
- cartoon
- 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
- 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
Abstract
The invention belongs to picture appraisal technical fields, disclose a kind of compound degraded image quality evaluating method decomposed based on cartoon texture and system, first, the conspicuousness marginal texture information of image and grain details information are effectively distinguished using cartoon texture decomposition algorithm；Secondly, fuzzy evaluation is carried out to the cartoon component of image using characteristic similarity algorithm, obtains the fuzzy evaluation factor；Then, noise strength factor is obtained using texture component；After consistent correction is carried out to the two, with reference to human-eye visual characteristic, weight parameter is provided, finally obtains the Environmental Evaluation Model of the compound degraded image of fuzzy noise.The present invention has higher subjective and objective consistency for the fuzzy and compound degraded image of noise in LIVEMD image libraries, the evaluation result of algorithm.
Description
Technical field
The invention belongs to picture appraisal technical field more particularly to a kind of compound degraded images decomposed based on cartoon texture
Quality evaluating method and system.
Background technology
At present, the prior art commonly used in the trade is such：
Image is the main source that people obtain information, with smart mobile phone, digital camera it is universal, all the time largely
Image be acquired, the Internet technology of rapid development causes the transmission of these images to facilitate with sharing quick.In image
It obtains, during transmission, processing and display, some inevitable disturbing factors frequently can lead to the decline of picture quality, such as
Electronic noise, shake obscure, loss of data etc. caused by compression, and therefore, it is accurately to recognize to carry out reliable quality evaluation to image
Know and preferably utilize the premise of image.In recent years, image quality evaluation (Image Quality Assessment, IQA) has been
As one important research direction of image processing field and research hotspot.Image quality evaluating method generally can be divided into subjective assessment
Method and objective evaluation, subjective assessment carry out picture quality subjective marking by several observers, and common marking mechanism has flat
Equal subjective scores method (Mean Opinion Score, MOS) or the average subjective scores method of difference (Differential Mean
Opinion Score,DMOS).Although subjective estimate method is the most accurate description to picture quality, heavy workload takes
It is long, it is very difficult to carry out practical application；Objective evaluation refers to by building mathematical model, is calculated by computer and subjective assessment
It is worth more consistent image quality index, objective evaluation application is convenient, efficient, is that people carry out the main of image quality evaluation
Means.
According to the different demands to reference picture, Objective image quality evaluation method can be divided into：Method (Full- is referred to entirely
Reference, FR), partly with reference to method (Reduced-Reference, RR) and without with reference to method (No-Reference, NR).Entirely
The appearance of reference method structural similarity (Structure SIMilarity, SSIM), is greatly promoted image quality evaluation
Progress so that algorithm from traditional Y-PSNR (Peak Signal Noise Ratio, PSNR) based on pixel,
Mean square error (Mean Square Error, MSE) etc. is changed into structure-based algorithm, has emerged a large amount of outstanding algorithm,
Such as the various improved models of SSIM algorithms：SSIM based on information weight, characteristic similarity model (Feature
SIMilarity, FSIM), the Similarity Model DASM of anisotropic structure etc.；Consider the model of human-eye visual characteristic：Based on figure
Model as salient region, model based on area-of-interest etc..In the application process of Objective image quality evaluation method,
The missing of reference picture is be difficult to avoid that the problem of, thus, it is only required to which the part of partial reference image information is with reference to method and not
Need to refer to image without with reference to evaluation assessment, also referred to as blind image quality evaluation (Blind IQA, BIQA) is increasingly becoming research heat
Point.Wherein, great representative algorithm has：It is empty with reference to evaluation, blind/non-reference picture based on the part that picture structure pattern degrades
Domain quality evaluation (Blind/Referenceless Image Spatial QUality Evaluator, BRISQUE), nature
Image quality evaluation (Natural Image Quality Evaluator, NIQE), the quality evaluation based on spatial frequency spectrum and entropy
(Spatial-Spectral Entropy-based Quality index, SSEQ) etc..
With the continuous development that image quality evaluation is studied, researcher is dedicated to inquiring into deeper into, more complicated image matter
Measure evaluation problem.The image in University of Texas's Jane Austen branch school and the compound drop of video engineering experiment room successively issue in recent years
Matter image library LIVEMD, the true degraded image library ChallengeDB in field, high dynamic range images library HDRdatabase, constantly
Existing image quality evaluation algorithm is challenged, explores new research direction.There is document to propose the matter for including compound degraded image
Critical issue urgently to be resolved hurrily in seven big challenges and image quality evaluation evolution including amount evaluation problem.
In conclusion problem of the existing technology is：
(1) for it is compound degrade in the compound degradation problems of fuzzy noise carry out evaluation algorithms and have at following 2 points：
First, obscure and noise caused by quality decline be it is most common in image acquisition, transmission and processing procedure and
Be most difficult to avoid, and image is easy to be influenced by the two simultaneously, thus both explore it is compound degrade with it is practical should
With value；And the enlightenment that the prior art has not provided；
Secondly, degrade relative to fuzzy and the compound of JPEG, JPEG2000, the figure of fuzzy modeling of relying with noise to algorithm
As feature influence in a way have acting in opposition, be overlapped mutually also increasingly complex with interference mechanism.
The prior art does not propose a kind of compound degraded image evaluation model of general fuzzy noise, to image card
After logical texture decomposes, the degree that degrades respectively to fuzzy and noise is evaluated, and efficiently avoids interfering with each other for the two；
And the prior art is estimated not over collecting strategy and obtain image final mass；Cause subjective and objective consistency not
By force, it is impossible to which compound degraded image quality is accurately evaluated.
Solve the difficulty and meaning of above-mentioned technical problem：
The present invention proposes a kind of compound degraded image evaluation model of general fuzzy noise, and cartoon texture is being carried out to image
After decomposition, the degree that degrades respectively to fuzzy and noise is evaluated, and efficiently avoids interfering with each other for the two；Finally, pass through
Collect strategy and obtain image final mass and estimate.It is higher that experiment in LIVEMD image libraries shows that inventive algorithm has
Subjective and objective consistency can accurately evaluate compound degraded image quality.
Invention content
In view of the problems of the existing technology, the present invention provides a kind of compound degraded images decomposed based on cartoon texture
Quality evaluating method and system.
The invention is realized in this way a kind of compound degraded image quality evaluating method decomposed based on cartoon texture, is：
First, the conspicuousness marginal texture information of image and grain details information are carried out using cartoon texture decomposition algorithm
Effectively distinguish；
Secondly, fuzzy evaluation is carried out to the cartoon component of image using characteristic similarity algorithm, obtains the fuzzy evaluation factor；
Then, noise strength factor is obtained using texture component；After consistent correction is carried out to the two, regarded with reference to human eye
Feel characteristic, provide weight parameter, finally obtain the Environmental Evaluation Model of the compound degraded image of fuzzy noise；
CTDMDI=ω1·FSIMmap+ω2·NLmap；
In formula, FSIMmap, NLmapRespectively the fuzzy evaluation index of image, noise intensity index are in subjective assessment DMOS domains
Mapping, ω1=0.7, ω2=0.3, respectively the linear weighted function value of fuzzy evaluation index and noise rating index.
Further, the compound degraded image quality evaluating method decomposed based on cartoon texture is specifically included：
1) cartoon texture decomposes；The piecewise smooth part of image and detail textures part are distinguished, by picture breakdown
For cartoon component and texture component；
2) fuzzy evaluation based on cartoon component；Fog-level is carried out to the cartoon component of compound degraded image using FSIM
Evaluation；
3) the noise intensity evaluation based on texture component；Based on the weak texture information of image noise intensity evaluation and test algorithm come pair
The texture component of compound degraded image carries out noise intensity evaluation；
4) collect the design of strategy, utilize fuzzy and noise image the subjective evaluation index DMOS in LIVE2 image libraries
As reference, according to fitting formula, fuzzy factor and noise factor are respectively mapped to subjective assessment domain by fitting.
Further, fuzzy evaluation of the step 2) based on cartoon component；Including：
The phase equalization value of each pixel is calculated, respectively obtains the phase equalization figure PC of two images1And PC2；
The Grad of each pixel is calculated, obtains the gradient matrix G of two images1And G2；
Calculate the phase portrait figure S of reference picture and degraded imagePCWith gradient similitude figure SG；
In formula (1) and formula (2), T1、T2For normal number, for increasing SPCAnd SGStability, value T1=0.85, T2
=160, value is codetermined by the dynamic range and experimental debugging of PC, G value respectively；
Calculate the similitude figure S of reference picture and degraded imageL；
SL=SPC·SG(3)；
Using the phase equalization index of expression image information relative importance as weight, the quality for finally establishing image is commented
Valency model；
In formula (4), PCm(x)=max (PC1(x),PC2(x))。
Further, image noise evaluation of the step 3) based on texture component includes：
To additive white Gaussian noise, it is expressed as：yi=zi+ni, wherein, yiRepresent i-th of noisy image block region, ziIt is
I noise-free picture block region, niRepresent noise component(s)；
Using principal component analysis, obtain：
Wherein, ∑ y is noisy image region yiCovariance matrix, ∑ z is noise-free picture region ziCovariance matrix,Represent noise variance, λmin(∑ y), λmin(∑ z) is respectively the minimal eigenvalue of corresponding covariance matrix.
For the redundancy of natural image, covariance matrix is low-rank matrix, λmin(∑ z) regards 0 as, then noise intensity
(Noise Level, NL) is obtained by following calculate：
Further, the step 4) is collected tactful design and is included：
In formula (7), q is multiple image objective evaluation indicator vector；Q is the vector being mapped to behind subjective assessment domain；εi(i
=1,2,3,4) it is fitting parameter；
It is as follows：
FSIM algorithms and NL algorithms are utilized respectively to 145 width blurred pictures in LIVE2 image libraries, 145 panel heights this noise patterns
As carrying out quality evaluation, objective evaluation vector q is obtainedblur, qnoise；
With reference to the DMOS provided in LIVE2 librariesblurAnd DMOSnoiseSubjective assessment vector, is fitted using formula (7), point
The subjective and objective fitting parameter ε of FSIM, NL are not obtainedi(i=1,2,3,4)；
Using fitting parameter, the NL of the FSIM indexs of the compound degraded image cartoon component of fuzzy noise and texture component is referred to
Mark is respectively mapped to subjective assessment domain, index value FSIM after being fittedmapAnd NLmap, at this point, the two has linear consistency；
The visual characteristic of the information of low level is more concerned with when observing image based on human eye, in linear weighted model more
Fuzzy indicator is laid particular emphasis on, establishes the compound degraded image Environmental Evaluation Model (Cartoon-Texture decomposed based on cartoon texture
Decomposition based Multi-Distortion Index, CTDMDI)：
CTDMDI=ω1·FSIMmap+ω2·NLmap(8)；
In formula (8), ω1=0.7, ω2=0.3, the respectively linear weighted function of fuzzy evaluation index and noise rating index
Value.
Another object of the present invention is to provide the compound degraded image matter decomposed described in a kind of realize based on cartoon texture
Measure the computer program of evaluation method.
Another object of the present invention is to provide a kind of to realize the compound drop decomposed based on cartoon texture described in claim
The information data processing terminal of matter image quality evaluating method.
Another object of the present invention is to provide a kind of computer readable storage medium, including instruction, when it is in computer
During upper operation so that computer performs the compound degraded image quality evaluating method decomposed based on cartoon texture.
Another object of the present invention is to provide a kind of compound degraded image quality evaluation system decomposed based on cartoon texture
System, including：
Cartoon texture resolving cell；For the piecewise smooth part of image and detail textures part to be distinguished, will scheme
As being decomposed into cartoon component and texture component；
Fuzzy evaluation unit based on cartoon component；Fuzzy journey is carried out to the cartoon component of compound degraded image using FSIM
Degree evaluation；
Noise intensity evaluation unit based on texture component；It is calculated for being evaluated and tested based on the noise intensity of the weak texture information of image
Method to carry out noise intensity evaluation to the texture component of compound degraded image；
Collect tactful design cell, utilize fuzzy and noise image the subjective evaluation index DMOS in LIVE2 image libraries
As reference, according to fitting formula, fuzzy factor and noise factor are respectively mapped to subjective assessment domain by fitting.
It is a kind of equipped with the compound degraded image decomposed based on cartoon texture another object of the present invention is to provide
The information data processing terminal of QA system.
In conclusion advantages of the present invention and good effect are：
The present invention is in order to improve the accuracy of compound degraded image quality evaluation, using cartoon texture decomposition algorithm by image
Cartoon component and texture component are decomposed into, image fuzzy factor is obtained from the conspicuousness marginal information in cartoon component, from texture
Component obtains the noise strength factor of image, with reference to strategy is collected, has constructed effective compound degraded image quality evaluation mould
Type.Experiment shows that the evaluation result of algorithm is with higher for the fuzzy and noise compound degraded image in LIVEMD image libraries
Subjective and objective consistency.
Experiment in LIVEMD image libraries shows that the present invention can accurately comment compound degraded image quality
Valency.
Description of the drawings
Fig. 1 is the compound degraded image quality evaluating method flow provided in an embodiment of the present invention decomposed based on cartoon texture
Figure.
Fig. 2 be it is provided in an embodiment of the present invention using algorithm in document to " lakebuilding " in LIVEMD image libraries
Image carries out the result figure of cartoon texture decomposition.
In figure：(a), original image；(b) cartoon component；(c) texture component.
Fig. 3 is picture quality objective evaluation algorithm PSNR, SSIM, FSIM, GMSD provided in an embodiment of the present invention,
BRISQUE, NIQE and inventive algorithm and the fitting scatter plot of subjective assessment DMOS values.
Fig. 4 is picture quality objective evaluation algorithm IL-NIQE, Pro., GWH-GLBP provided in an embodiment of the present invention and sheet
Invention algorithm and the fitting scatter plot of subjective assessment DMOS values.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
The prior art does not propose a kind of compound degraded image evaluation model of general fuzzy noise, to image card
After logical texture decomposes, the degree that degrades respectively to fuzzy and noise is evaluated, and efficiently avoids interfering with each other for the two；
And the prior art is estimated not over collecting strategy and obtain image final mass；Cause subjective and objective consistency not
By force, it is impossible to which compound degraded image quality is accurately evaluated.
With reference to concrete analysis, the invention will be further described.
1st, the compound degraded image quality evaluation decomposed based on cartoon texture
The evaluation model that the present invention establishes mainly has 4 parts：1) cartoon texture decomposes；2) it is commented based on the fuzzy of cartoon component
Valency；3) the noise intensity evaluation based on texture component；4) collect the design of strategy.
Fig. 1 is the flow chart of the present invention.
First, the conspicuousness marginal texture information of image and grain details information are carried out using cartoon texture decomposition algorithm
Effectively distinguish；Secondly, fuzzy evaluation is carried out to the cartoon component of image using characteristic similarity algorithm, obtain fuzzy evaluation because
Son；Then, noise strength factor is obtained using texture component；After consistent correction is carried out to the two, with reference to human eye vision spy
Property, weight parameter is provided, finally obtains the quality evaluation index of the compound degraded image of fuzzy noise.
1.1 cartoon textures decompose
1.1.1 the basic thought that cartoon texture decomposes
The decomposition of cartoon texture is the very significant concept in one, picture breakdown field, and main thought is by point of image
Section smooth part and detail textures part distinguish, i.e., are cartoon component and texture component by picture breakdown, with preferably right
Image carries out subsequent processing.This picture breakdown mode be widely used in compression of images, image analysis, image super-resolution
The fields such as rate reconstruction and computer vision.
Fig. 2 carries out " lakebuilding " image in LIVEMD image libraries cartoon texture point using algorithm in document
The result of solution.
1.1.2 the separation with noise is obscured
It is blurred image to evaluate the conspicuousness edge for being generally basede on image, and the flat of image is often utilized to the evaluation of noise
Skating area domain, the less region of structural information, therefore, carries out cartoon texture by the compound degraded image of fuzzy and noise herein in other words
Decompose, purpose with using cartoon texture decompose characteristic, by the conspicuousness marginal texture information of image with including the thin of noise
Section texture information distinguishes.
Degrading by reference picture and by fuzzy and noise is compound " the cartoon textures of lakebuilding " images decomposes
The comparison of figure is it is found that the cartoon component that decomposition obtains mainly remains the conspicuousness structural information of image, substantially not by noise shadow
It rings；By comparing it is found that noise is broken down into as detailed information in texture component.
The 1.2 image fuzzy evaluations based on cartoon component
To the continuous of HVS, research shows that, human eye mainly obtains the understanding of image by low level feature, has document to carry
One kind is gone out using phase equalization as main feature, using gradient as the characteristic similarity image quality evaluation index of supplemental characteristic
(Feature SIMmilarity,FSIM).For known reference image and degraded image, its step are as follows：
1) the phase equalization value of each pixel is calculated, respectively obtains the phase equalization figure PC of two images1And PC2。
2) Grad of each pixel is calculated, obtains the gradient matrix G of two images1And G2。
3) the phase portrait figure S of reference picture and degraded image is calculatedPCWith gradient similitude figure SG。
In formula (1) and formula (2), T1、T2For normal number, for increasing SPCAnd SGStability, value T1=0.85, T2
=160, value is codetermined by the dynamic range and experimental debugging of PC, G value respectively.
4) the similitude figure S of reference picture and degraded image is calculatedL。
SL=SPC·SG (3).
In view of human-eye visual characteristic, the phase equalization index of image information relative importance will be represented as weight,
The final Environmental Evaluation Model for establishing image.
In formula (4), PCm(x)=max (PC1(x),PC2(x))。
By algorithm principle and step it is found that FSIM indexs more focus on the low level information of image, using FSIM to compound drop
The cartoon component of matter image carries out fog-level evaluation, can ignore the influence of detailed information to a certain extent, obtain compared with subject to
The true vague intensity factor.
1.3 image noise evaluations based on texture component
It is continued to bring out for the image noise evaluation algorithm of additive white Gaussian noise, outstanding algorithm therein has been able to standard
True acquisition noise variance, the present invention are calculated using a kind of noise intensity evaluation and test based on the weak texture information of image proposed in document
Method to carry out noise intensity evaluation to the texture component of compound degraded image, and algorithm principle and step are as follows：
1) it to additive white Gaussian noise, is generally expressed as below：yi=zi+ni, wherein, yiRepresent i-th of noisy image block
Region, ziFor i-th of noise-free picture block region, niRepresent noise component(s).
2) it using principal component analysis, can obtain drawing a conclusion：
Wherein, ∑ y is noisy image region yiCovariance matrix, ∑ z is noise-free picture region ziCovariance matrix,Represent noise variance, λmin(∑ y), λmin(∑ z) is respectively the minimal eigenvalue of corresponding covariance matrix.
3) since natural image has redundancy, covariance matrix is often low-rank matrix, therefore, λmin(∑ z) can be with
Regard 0 as, then noise intensity (Noise Level, NL) can be obtained by following calculate：
4) texture component of the invention can regard a weak texture image as, and image-region includes a large amount of flat site,
Low-rank image-region is belonged to, the noise intensity in texture component can be effectively obtained using this algorithm.
1.4 collect strategy
The vague intensity factor of cartoon component and the noise strength factor of texture component do not have linear dependence, for structure
Build the final mass evaluation model of compound degraded image, it is necessary first to which consistent correction is carried out to the two.The present invention utilizes LIVE2
Fuzzy and noise image subjective evaluation index (DMOS) in image library as reference, will according to the fitting formula in document
Fuzzy factor and noise factor are respectively mapped to subjective assessment domain by fitting.
In formula (7), q is multiple image objective evaluation indicator vector；Q is the vector being mapped to behind subjective assessment domain；εi(i
=1,2,3,4) it is fitting parameter.
Realize that step is as follows：
1) FSIM algorithms and NL algorithms are utilized respectively to 145 width blurred pictures, 145 width Gaussian noises in LIVE2 image libraries
Image carries out quality evaluation, obtains objective evaluation vector qblur, qnoise。
2) with reference to the DMOS provided in LIVE2 librariesblurAnd DMOSnoiseSubjective assessment vector, is fitted using formula (7),
Respectively obtain the subjective and objective fitting parameter ε of FSIM, NLi(i=1,2,3,4).
3) using fitting parameter, by the NL of the FSIM indexs of the compound degraded image cartoon component of fuzzy noise and texture component
Index is respectively mapped to subjective assessment domain, index value FSIM after being fittedmapAnd NLmap, at this point, the two has linearly unanimously
Property.
4) visual characteristic of the information of low level is more concerned with when observing image based on human eye, in linear weighted model
Fuzzy indicator is more focused on, establishes the compound degraded image Environmental Evaluation Model (Cartoon- decomposed based on cartoon texture
Texture DecompositionbasedMulti-Distortion Index, CTDMDI)：
CTDMDI=ω1·FSIMmap+ω2·NLmap (8).
In formula (8), ω1=0.7, ω2=0.3, the respectively linear weighted function of fuzzy evaluation index and noise rating index
Value.In formula, FSIMmap, NLmapRespectively the fuzzy evaluation index of image, noise intensity index are reflected subjective assessment DMOS domains
It penetrates.
The theoretical foundation of its value to be found in the research to human visual system, human eye in perceptual image, depending on
Focus is felt often in the conspicuousness marginal portion of image, it is therefore, higher to the edge blurry susceptibility of image, and to image smoothing
The noise relative perceptual of region and smooth region is weaker, is given in experiment with analysis part and chooses optimal weights using experiment
Tables of data, from confirm on the other hand human eye vision conspicuousness theory.
The embodiment of the present invention provides a kind of compound degraded image QA system decomposed based on cartoon texture, including：
Cartoon texture resolving cell；For the piecewise smooth part of image and detail textures part to be distinguished, will scheme
As being decomposed into cartoon component and texture component；
Fuzzy evaluation unit based on cartoon component；Fuzzy journey is carried out to the cartoon component of compound degraded image using FSIM
Degree evaluation；
Noise intensity evaluation unit based on texture component；It is calculated for being evaluated and tested based on the noise intensity of the weak texture information of image
Method to carry out noise intensity evaluation to the texture component of compound degraded image；
Collect tactful design cell, utilize fuzzy and noise image the subjective evaluation index DMOS in LIVE2 image libraries
As reference, according to fitting formula, fuzzy factor and noise factor are respectively mapped to subjective assessment domain by fitting.
With reference to experiment, the invention will be further described with analysis.
2 experiments and analysis
2.1 compound degraded image library LIVEMD
First compound degraded image quality assessment database has been issued in LIVE laboratories in 2013, and 15 width marks are included in library
Quasi- reference picture has carried out every width reference picture two kinds of compound operations that degrade respectively：1) image by it is fuzzy degrade after, it is right
It carries out JPGE compressions, obtains compound degraded image；2) image by it is fuzzy degrade after, add white Gaussian noise, obtain
Compound degraded image.Each type that degrades has 3 grades, and 225 width have been obtained and have obscured the compound degraded image of JPEG compression and 225
The compound degraded image of width fuzzy noise.The DMOS subjectivity marking values of every width degraded image are given in library, DMOS values are smaller, image
Quality is better, conversely, picture quality is poorer.In an experiment, it has been carried out herein for 225 width fuzzy noise degraded images subjective and objective
Consistency is tested.
2.2 optimal weights choose experiment
Experiment, respectively to increase FSIM algorithm index weights, reduces NL algorithm index weights since average weighted thinking
Direction and reduction FSIM algorithm index weights, increase NL algorithm index weights both directions are tested, test step a length of 0.1.
Experimental result is as shown in table 1, and for the accuracy of experiment, step-length is set as 0.05 being likely to occur at turning point：
1 weight parameter test experiments data of table (add * parts for this paper weights and corresponding subjective and objective coincident indicator)
As shown in Table 1, weight is set as：ω1=0.7, ω2=0.3, existing theoretical foundation also complies with experiment test knot
Fruit.
2.3 algorithm evaluation parameters
Since human eye is the final recipient of image, Objective image quality evaluation index is to what extent subjective with human eye
Judge that consistent is conventional method whether evaluating its validity.Common 5 subjective and objective indicator consilience evaluation parameters include：
1) Pearson's linearly dependent coefficient (Pearson Linear Correlation Coefficient, PLCC)；2) it is mapped to master
See the linearly dependent coefficient (PLCC after mapping, PLCC (mapping)) behind evaluation index domain；3) Spearman order phase
Relationship number (Spearman Rank-Order Correlation Coefficient, SROCC)；4) mean square error (Root
Mean Square Error,RMSE)；5) absolute average error (Mean Absolute Error, MAE).Wherein, PLCC,
RMSE and MAE is used for the accuracy of evaluation algorithms, and SROCC is used for the monotonicity of evaluation algorithms.One outstanding objective image matter
Measure evaluation algorithms should have higher subjective and objective consistency, i.e. PLCC, PLCC (mapping), SROCC values close to 1, and with subjectivity
The error of evaluation is smaller, i.e. RMSE, MAE value is smaller.
2.4 experiments and analysis
In order to better illustrate the validity and accuracy that this paper algorithms evaluate compound degraded image, this section evaluates it
Performance has carried out subjective and objective consistency check, and is compared with existing multiple outstanding algorithms.Since some algorithm is base
In trained, image library is usually divided into 80% training sample set and 20% test sample collection, and sample size difference
Huge test sample collection is difficult to obtain reliable comparing result.Therefore, two groups of contrast experiments have been carried out altogether herein：
Contrast experiment 1：Test set is the 225 compound degraded images of width fuzzy noise in LIVEMD image libraries, compares algorithm
Including：PSNR, SSIM, FSIM, GMSD, BRISQUE, SSEQ and NIQE.Wherein, PSNR, SSIM, FSIM, GMSD and the present invention
Algorithm CTDMDI is FR evaluation algorithms, and BRISQUE, SSEQ and NIQE are NR evaluation algorithms.As shown in Figure 3.
Contrast experiment 2：Test set is the 45 compound degraded images of width fuzzy noise randomly selected, i.e. 20% sample of image library
This, comparison algorithm includes：Based on trained IL-NIQE algorithms [20], Pro. algorithms [21] and GWH-GLBP algorithms.Such as Fig. 4 institutes
Show.
All algorithms are applied to corresponding test set, corresponding image objective evaluation index are obtained, using in image library
The subjective evaluation index DMOS provided, is fitted using Matlab Fitting Toolbox,
As a result as shown in Figure 3,4, subjective and objective compliance evaluation parameter is shown in Table 2.
2 this paper algorithms of table and comparison algorithm evaluation performance compare (the optimal overstriking in each index is shown)
By two groups of contrast experiments in Fig. 3 and table 2 it is found that FR evaluation algorithms are commented on the whole better than NR evaluation algorithms in NR
In valency algorithm, BRISQE, SSEQ algorithms show excellent BRISQUE in non-composite degraded image quality evaluation], SSEQ, and it is right
The compound degraded image of fuzzy noise then cannot be evaluated reliably substantially, the reason is that the feature that uses of algorithm modeling fuzzy and
To two opposite direction changes under the influence of noise, when the two exists simultaneously, it is overlapped mutually and interference leads to algorithm base
Originally role of evaluation is lost, there is document] detailed explanation has been carried out to this.By taking BRISQE as an example, algorithm is characterized in for modeling
MSCN (Mean Subtracted Contrast Normalized), calculation formula is：
In formula (9), I (i, j) is original image,To remove image after equalization weights.μ (i, j), σ (i, j) are respectively
Gauss weighted average and variance for the image block centered on pixel (i, j).Under noise and Fuzzy Influence, the change of MSCN
Change the both direction for tending to opposite.When noise is with fuzzy exist simultaneously, the variation of MSCN loses under the two collective effect
The ability of image deterioration degree is characterized, the subjective and objective consistency of final mass evaluation is poor, as shown in Figure 3.SSEQ algorithms are built
Modular character there is also it is similary the problem of so that both algorithms are only single degraded image and carry out accurate quality evaluation, and
It cannot be used for the compound degraded image quality evaluation of fuzzy noise.
This paper algorithms are decomposed by cartoon texture, efficiently avoid it is fuzzy and noise on image feature influence each other,
Accurate evaluation has been carried out to picture quality.From algorithm evaluation parameter, compared to ten comparison algorithms, accuracy
It is all greatly improved with monotonicity.It compares in algorithm, NR evaluation index Pro. and GWH-GLBP has certain with this paper algorithms
Comparativity, but its modeling is based on trained, and algorithm complexity is higher, and for different image libraries, and algorithm is required for weight
It is newly trained, can just obtain optimal evaluation result.In contrast, this paper algorithms are feature baseds, do not need to carry out a large amount of
Training, it is possible to more convenient in practical applications applied to all compound degraded images of fuzzy noise.
With reference to result, the invention will be further described.
Difficult, this problem of visual perception complexity for the compound degraded image quality evaluation modeling of fuzzy and noise, herein
Propose a kind of compound degraded image quality evaluation algorithm decomposed based on cartoon texture.Algorithm is decomposed using cartoon texture will figure
The detail textures data separation of the big profile information such as the conspicuousness structure of picture, edge and image comes, and is respectively used to fog-level
With the evaluation of noise intensity, meanwhile, with reference to human-eye visual characteristic design collect strategy, obtain best evaluation effect.
It is in LIVEMD image libraries the experimental results showed that, this paper algorithms and subjective evaluation result have a higher consistency, evaluation it is accurate
Property with monotonicity better than comparison algorithm.
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or its arbitrary combination real
It is existing.Entirely or partly realized in the form of a computer program product when using, the computer program product include one or
Multiple computer instructions.When loading on computers or performing the computer program instructions, entirely or partly generate according to
Flow or function described in the embodiment of the present invention.The computer can be all-purpose computer, special purpose computer, computer network
Network or other programmable devices.The computer instruction can be stored in a computer-readable storage medium or from one
Computer readable storage medium is transmitted to another computer readable storage medium, for example, the computer instruction can be from one
A web-site, computer, server or data center pass through wired (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)
Or wireless (such as infrared, wireless, microwave etc.) mode is carried out to another web-site, computer, server or data center
Transmission).The computer read/write memory medium can be that any usable medium that computer can access either includes one
The data storage devices such as server, the data center that a or multiple usable mediums integrate.The usable medium can be magnetic Jie
Matter, (for example, floppy disk, hard disk, tape), optical medium (for example, DVD) or semiconductor medium (such as solid state disk Solid
State Disk (SSD)) etc..
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement made within refreshing and principle etc., should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of compound degraded image quality evaluating method decomposed based on cartoon texture, which is characterized in that described to be based on cartoon
Texture decompose compound degraded image quality evaluating method be：
First, the conspicuousness marginal texture information of image and grain details information are carried out using cartoon texture decomposition algorithm effective
It distinguishes；
Secondly, fuzzy evaluation is carried out to the cartoon component of image using characteristic similarity algorithm, obtains the fuzzy evaluation factor；
Then, noise strength factor is obtained using texture component；After consistent correction is carried out to the two, with reference to human eye vision spy
Property, weight parameter is provided, finally obtains the Environmental Evaluation Model of the compound degraded image of fuzzy noise；
CTDMDI=ω1·FSIMmap+ω2·NLmap；
In formula, FSIMmap, NLmapRespectively the fuzzy evaluation index of image, noise intensity index are reflected subjective assessment DMOS domains
It penetrates, ω1=0.7, ω2=0.3, respectively the linear weighted function value of fuzzy evaluation index and noise rating index.
2. the compound degraded image quality evaluating method decomposed as described in claim 1 in cartoon texture, which is characterized in that institute
The compound degraded image quality evaluating method decomposed based on cartoon texture is stated to specifically include：
1) cartoon texture decomposes；The piecewise smooth part of image and detail textures part are distinguished, are card by picture breakdown
Reduction of fractions to a common denominator amount and texture component；
2) fuzzy evaluation based on cartoon component；Fog-level is carried out using FSIM to the cartoon component of compound degraded image to comment
Valency；
3) the noise intensity evaluation based on texture component；Noise intensity based on the weak texture information of image evaluates and tests algorithm come to compound
The texture component of degraded image carries out noise intensity evaluation；
4) collect the design of strategy, by the use of the fuzzy and noise image in LIVE2 image libraries subjective evaluation index DMOS as
With reference to fuzzy factor and noise factor are respectively mapped to subjective assessment domain by foundation fitting formula by fitting.
3. the compound degraded image quality evaluating method decomposed as claimed in claim 2 in cartoon texture, which is characterized in that institute
State fuzzy evaluation of the step 2) based on cartoon component；Including：
The phase equalization value of each pixel is calculated, respectively obtains the phase equalization figure PC of two images1And PC2；
The Grad of each pixel is calculated, obtains the gradient matrix G of two images1And G2；
Calculate the phase portrait figure S of reference picture and degraded imagePCWith gradient similitude figure SG；
In formula (1) and formula (2), T1、T2For normal number, for increasing SPCAnd SGStability, value T1=0.85, T2=160,
Its value is codetermined by the dynamic range and experimental debugging of PC, G value respectively；
Calculate the similitude figure S of reference picture and degraded imageL；
SL=SPC·SG(3)；
Using the phase equalization index for representing image information relative importance as weight, the quality evaluation mould of image is finally established
Type；
In formula (4), PCm(x)=max (PC1(x),PC2(x))。
4. the compound degraded image quality evaluating method decomposed as claimed in claim 2 in cartoon texture, which is characterized in that institute
Image noise evaluation of the step 3) based on texture component is stated to include：
To additive white Gaussian noise, it is expressed as：yi=zi+ni, wherein, yiRepresent i-th of noisy image block region, ziIt is i-th
Noise-free picture block region, niRepresent noise component(s)；
Using principal component analysis, obtain：
Wherein, ∑yIt is noisy image region yiCovariance matrix, ∑zIt is noise-free picture region ziCovariance matrix,Table
Show noise variance, λmin(∑y), λmin(∑z) it is respectively the minimal eigenvalue for corresponding to covariance matrix.
For the redundancy of natural image, covariance matrix is low-rank matrix, λmin(∑z) regard 0 as, then noise intensity (Noise
Level, NL) it is obtained by following calculate：
5. the compound degraded image quality evaluating method decomposed as claimed in claim 2 based on cartoon texture, which is characterized in that
The design that the step 4) collects strategy includes：
In formula (7), q is multiple image objective evaluation indicator vector；Q is the vector being mapped to behind subjective assessment domain；εi(i=1,2,
3,4) it is fitting parameter；
It is as follows：
Be utilized respectively FSIM algorithms and NL algorithms to 145 width blurred pictures in LIVE2 image libraries, this noise image of 145 panel heights into
Row quality evaluation obtains objective evaluation vector qblur, qnoise；
With reference to the DMOS provided in LIVE2 librariesblurAnd DMOSnoiseSubjective assessment vector, is fitted using formula (7), respectively
To the subjective and objective fitting parameter ε of FSIM, NLi(i=1,2,3,4)；
Using fitting parameter, by the FSIM indexs of the compound degraded image cartoon component of fuzzy noise and the NL indexs point of texture component
It is not mapped to subjective assessment domain, index value FSIM after being fittedmapAnd NLmap, at this point, the two has linear consistency；
The visual characteristic of the information of low level is more concerned with when observing image based on human eye, is more stressed in linear weighted model
In fuzzy indicator, the compound degraded image Environmental Evaluation Model decomposed based on cartoon texture is established：
CTDMDI=ω1·FSIMmap+ω2·NLmap(8)；
In formula (8), ω1=0.7, ω2=0.3, respectively the linear weighted function value of fuzzy evaluation index and noise rating index.
6. a kind of realize the compound degraded image quality evaluation decomposed described in Claims 1 to 5 any one based on cartoon texture
The computer program of method.
7. a kind of realize the compound degraded image quality evaluation decomposed described in Claims 1 to 5 any one based on cartoon texture
The information data processing terminal of method.
8. a kind of computer readable storage medium, including instructing, when run on a computer so that computer is performed as weighed
Profit requires the compound degraded image quality evaluating method decomposed based on cartoon texture described in 1-5 any one.
9. it is a kind of as described in claim 1 based on cartoon texture decompose compound degraded image quality evaluating method based on card
The compound degraded image QA system that logical texture decomposes, which is characterized in that the compound drop decomposed based on cartoon texture
Matter image quality evaluation system includes：
Cartoon texture resolving cell；For the piecewise smooth part of image and detail textures part to be distinguished, by image point
It solves as cartoon component and texture component；
Fuzzy evaluation unit based on cartoon component；Fog-level is carried out using FSIM to the cartoon component of compound degraded image to comment
Valency；
Noise intensity evaluation unit based on texture component；For based on the noise intensity of the weak texture information of image evaluate and test algorithm come
Noise intensity evaluation is carried out to the texture component of compound degraded image；
Collect tactful design cell, by the use of the fuzzy and noise image in LIVE2 image libraries subjective evaluation index DMOS as
With reference to fuzzy factor and noise factor are respectively mapped to subjective assessment domain by foundation fitting formula by fitting.
10. a kind of letter equipped with the compound degraded image QA system decomposed described in claim 9 based on cartoon texture
Cease data processing terminal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810119012.4A CN108230325A (en) | 2018-02-06 | 2018-02-06 | The compound degraded image quality evaluating method and system decomposed based on cartoon texture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810119012.4A CN108230325A (en) | 2018-02-06 | 2018-02-06 | The compound degraded image quality evaluating method and system decomposed based on cartoon texture |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108230325A true CN108230325A (en) | 2018-06-29 |
Family
ID=62671020
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810119012.4A Pending CN108230325A (en) | 2018-02-06 | 2018-02-06 | The compound degraded image quality evaluating method and system decomposed based on cartoon texture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108230325A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961106A (en) * | 2018-07-27 | 2018-12-07 | 贵州省山地资源研究所 | KARST CAVES IN evaluation of tourism resources model and system, information data processing terminal |
| CN109754390A (en) * | 2018-12-11 | 2019-05-14 | 西北大学 | A kind of non-reference picture quality appraisement method based on mixing visual signature |
| CN110363716A (en) * | 2019-06-25 | 2019-10-22 | 北京工业大学 | One kind is generated based on condition and fights network combined degraded image high quality method for reconstructing |
| CN111161223A (en) * | 2019-12-13 | 2020-05-15 | 中国人民解放军国防科技大学 | Image interference degree evaluation method based on edge intensity similarity |
| CN111179245A (en) * | 2019-12-27 | 2020-05-19 | 成都中科创达软件有限公司 | Image quality detection method, device, electronic equipment and storage medium |
| CN111696064A (en) * | 2020-06-15 | 2020-09-22 | 北京金山云网络技术有限公司 | Image processing method, image processing device, electronic equipment and computer readable medium |
| CN111798421A (en) * | 2020-06-29 | 2020-10-20 | 浙江同善人工智能技术有限公司 | Image quality judging method, device and storage medium |
| CN114897885A (en) * | 2022-06-17 | 2022-08-12 | 北京东宇宏达科技有限公司 | Infrared image quality comprehensive evaluation system and method |
| CN117152023A (en) * | 2023-10-30 | 2023-12-01 | 浪潮软件科技有限公司 | Compressed image denoising method and device based on structure-texture decomposition |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107481221A (en) * | 2017-07-19 | 2017-12-15 | 天津大学 | Distorted image quality evaluating method is mixed with the full reference of cartoon rarefaction representation based on texture |
-
2018
- 2018-02-06 CN CN201810119012.4A patent/CN108230325A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107481221A (en) * | 2017-07-19 | 2017-12-15 | 天津大学 | Distorted image quality evaluating method is mixed with the full reference of cartoon rarefaction representation based on texture |
Non-Patent Citations (2)
| Title |
|---|
| FEIYAN ZHANG: "《Blind Quality Metric for Multidistortion Images Based on Cartoon and Texture Decomposition》", 《IEEE SIGNAL PROCESSING LETTERS》 * |
| LIN ZHANG: "《FSIM: A Feature Similarity Index for Image Quality Assessment》", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108961106A (en) * | 2018-07-27 | 2018-12-07 | 贵州省山地资源研究所 | KARST CAVES IN evaluation of tourism resources model and system, information data processing terminal |
| CN109754390A (en) * | 2018-12-11 | 2019-05-14 | 西北大学 | A kind of non-reference picture quality appraisement method based on mixing visual signature |
| CN109754390B (en) * | 2018-12-11 | 2023-04-07 | 西北大学 | No-reference image quality evaluation method based on mixed visual features |
| CN110363716B (en) * | 2019-06-25 | 2021-11-19 | 北京工业大学 | High-quality reconstruction method for generating confrontation network composite degraded image based on conditions |
| CN110363716A (en) * | 2019-06-25 | 2019-10-22 | 北京工业大学 | One kind is generated based on condition and fights network combined degraded image high quality method for reconstructing |
| CN111161223A (en) * | 2019-12-13 | 2020-05-15 | 中国人民解放军国防科技大学 | Image interference degree evaluation method based on edge intensity similarity |
| CN111179245A (en) * | 2019-12-27 | 2020-05-19 | 成都中科创达软件有限公司 | Image quality detection method, device, electronic equipment and storage medium |
| CN111179245B (en) * | 2019-12-27 | 2023-04-21 | 成都中科创达软件有限公司 | Image quality detection method, device, electronic equipment and storage medium |
| CN111696064A (en) * | 2020-06-15 | 2020-09-22 | 北京金山云网络技术有限公司 | Image processing method, image processing device, electronic equipment and computer readable medium |
| CN111696064B (en) * | 2020-06-15 | 2023-08-18 | 北京金山云网络技术有限公司 | Image processing method, device, electronic equipment and computer readable medium |
| CN111798421A (en) * | 2020-06-29 | 2020-10-20 | 浙江同善人工智能技术有限公司 | Image quality judging method, device and storage medium |
| CN114897885A (en) * | 2022-06-17 | 2022-08-12 | 北京东宇宏达科技有限公司 | Infrared image quality comprehensive evaluation system and method |
| CN117152023A (en) * | 2023-10-30 | 2023-12-01 | 浪潮软件科技有限公司 | Compressed image denoising method and device based on structure-texture decomposition |
| CN117152023B (en) * | 2023-10-30 | 2024-02-02 | 浪潮软件科技有限公司 | Compressed image denoising method and device based on structure-texture decomposition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230325A (en) | The compound degraded image quality evaluating method and system decomposed based on cartoon texture | |
| Yang et al. | A reference-free underwater image quality assessment metric in frequency domain | |
| Li et al. | No-reference image blur assessment based on discrete orthogonal moments | |
| Li et al. | Blind image quality assessment using statistical structural and luminance features | |
| Yang et al. | Perceptual quality assessment of screen content images | |
| Li et al. | Content-partitioned structural similarity index for image quality assessment | |
| Liu et al. | Image retargeting quality assessment | |
| Li et al. | No-reference quality assessment of deblocked images | |
| Shen et al. | Hybrid no-reference natural image quality assessment of noisy, blurry, JPEG2000, and JPEG images | |
| Liang et al. | No-reference perceptual image quality metric using gradient profiles for JPEG2000 | |
| CN107784651B (en) | Fuzzy image quality evaluation method based on fuzzy detection weighting | |
| CN109978854B (en) | Screen content image quality evaluation method based on edge and structural features | |
| Amirkhani et al. | An objective method to evaluate exemplar‐based inpainted images quality using Jaccard index | |
| CN101976444B (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
| Xu et al. | Hyperspectral image denoising using a spatial–spectral monte carlo sampling approach | |
| CN111415323B (en) | Image detection method and device and neural network training method and device | |
| CN106570862A (en) | Super-resolution reconstruction quality evaluation method and apparatus thereof | |
| Silvestre-Blanes | Structural similarity image quality reliability: Determining parameters and window size | |
| CN105894507B (en) | Image quality evaluating method based on amount of image information natural scene statistical nature | |
| Okarma | Extended hybrid image similarity–combined full-reference image quality metric linearly correlated with subjective scores | |
| Yang et al. | Image quality assessment based on the space similarity decomposition model | |
| Chi et al. | Blind tone mapped image quality assessment with image segmentation and visual perception | |
| Yang et al. | No‐reference image quality assessment via structural information fluctuation | |
| CN112132774A (en) | Quality evaluation method of tone mapping image | |
| Gao et al. | A content-based image quality metric |
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: 20180629 |