CN104954778B - Objective stereo image quality assessment method based on perception feature set - Google Patents
Objective stereo image quality assessment method based on perception feature set Download PDFInfo
- Publication number
- CN104954778B CN104954778B CN201510303868.3A CN201510303868A CN104954778B CN 104954778 B CN104954778 B CN 104954778B CN 201510303868 A CN201510303868 A CN 201510303868A CN 104954778 B CN104954778 B CN 104954778B
- Authority
- CN
- China
- Prior art keywords
- pixel
- designated
- dis
- org
- coordinate position
- 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.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an objective stereo image quality assessment method based on a perception feature set. According to the method, the distortion degree of a stereo image is measured according to the distortion degree of a visual perception feature map, a saliency map, a gradient map and an airspace just noticeable distortion map which are related to viewpoint perception quality are extracted, a disparity map related to third dimension and quality is extracted, distortion degrees of four perception feature maps are taken as features of the stereo image to create the perception feature set, a complicated human vision system is simulated according to a random forest machine learning algorithm, and feature parameters are fused. The objective stereo image quality assessment method has the advantages that change conditions of visual quality under the condition that the stereo image is influenced by various image processing and compressing methods can be reflected objectively, assessment performance of the method is not influenced by the content of the stereo image and the type of distortion, and conformity with subjective perception of human eyes is achieved.
Description
Technical field
The present invention relates to a kind of image quality evaluating method, more particularly, to a kind of stereo-picture based on perception feature set
Assessment method for encoding quality.
Background technology
It is three-dimensional video-frequency technology that stereo image quality evaluates (Stereo Image Quality Assessment, SIQA)
An important component.Stereoscopic image/video can inevitably be introduced in collection, storage, treatment and transmission and made an uproar
Sound or distortion, cause the decline of stereoscopic image/video quality, therefore, the evaluation to three-dimensional image/video quality is a needs
Research and the major issue for solving.Stereo image quality evaluation method is generally divided into subjective and objective two classes evaluation method.It is subjective
Evaluation method haves the shortcomings that time-consuming, laborious, high cost because of it, is difficult to realize and applies；And method for objectively evaluating is integrated
Algorithm, model, can quickly and easily obtain the objective quality of stereo-picture, without manual intervention, but its evaluation method is still
Prematurity, awaits further investigation, therefore method for objectively evaluating has turned into the emphasis of research.
In recent years, the research of stereo image quality objective evaluation achieves a series of achievements, can generally be divided into two
Class：The first kind：The evaluation method of plane picture is directly applied to the quality evaluation of stereo-picture, as You et al. will be classical
Plane picture quality evaluating method, such as PSNR, MS-SSIM, VIF are directly used in commenting for left view dot image and right visual point image
Valency, correspondence obtains the mass value of left view dot image and the mass value of right visual point image, then takes two average value conducts of mass value
The quality of stereo-picture, but stereo-picture makes a big difference with the quality evaluation of plane picture, depth perception distortion level
It is the key factor of the perceived quality for influenceing stereo-picture.Equations of The Second Kind：Added on the basis of plane picture quality evaluating method
Parallax information improves the evaluation model of stereo-picture, and such as Yang et al. adds the antipode figure of stereo-picture to carry out stereogram
As quality evaluation；And for example Benoit et al. is by depth information and plane picture quality evaluating method combining assessment stereo-picture matter
Amount；Hachicha et al. proposes the stereo image quality evaluation method that distortion is just perceived based on binocular, and uses binocular fusion
Relief change is weighed in competition, but, the third dimension of stereo-picture is the result that binocular fusion, binocular competition, binocular suppress,
Its mechanism is extremely complex, therefore relief measurement is extremely difficult, and this also causes these methods consistent with subjective perception
Property is relatively low.The above stereo image quality evaluation method is all to obtain global quality by local quality, or with linear or non-thread
The Feature fusion of property obtains the objective quality of stereo-picture, but human visual system is extremely complex system, these
Evaluation method all cannot the preferable complicated human visual system of simulation, cause to evaluate accuracy relatively low.
The content of the invention
The technical problems to be solved by the invention are to provide that a kind of stereo image quality based on perception feature set is objective to be commented
Valency method, it can effectively improve the correlation between objective evaluation result and subjective perception.
The present invention solve the technical scheme that is used of above-mentioned technical problem for：A kind of stereo-picture based on perception feature set
Assessment method for encoding quality, it is characterised in that comprise the following steps：
1. I is madeorgOriginal undistorted stereo-picture is represented, I is madedisThe stereo-picture of distortion to be evaluated is represented, will
IorgLeft view dot image be designated as Lorg, by IorgRight visual point image be designated as Rorg, by IdisLeft view dot image be designated as Ldis, will
IdisRight visual point image be designated as Rdis；
2. frequency modulation conspicuousness detection algorithm is used, L is obtainedorg、Rorg、LdisAnd RdisRespective notable figure, correspondence is designated as
WithThen calculateWithMean square error, be designated as
Equally, calculateWithMean square error, be designated as
Wherein, M represents IorgAnd IdisWidth, N represents IorgAnd IdisHeight, 1≤i≤M, 1≤j≤N,
RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
The pixel value of point,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle seat
Mark is set to the pixel value of the pixel of (i, j)；
3. horizontal Sobel operators are used, L is obtainedorg、Rorg、LdisAnd RdisRespective horizontal gradient figure, correspondence is designated asWithAnd vertical Sobel operators are used, obtain Lorg、Rorg、LdisAnd RdisRespective vertical ladder
Degree figure, correspondence is designated asWithThen basisWithObtain LorgGradient map, be designated as
WillMiddle coordinate position is designated as the pixel value of the pixel of (i, j) And root
According toWithObtain RorgGradient map, be designated asWillMiddle coordinate position is the pixel value of the pixel of (i, j)
It is designated as According toWithObtain LdisGradient map, be designated as
WillMiddle coordinate position is designated as the pixel value of the pixel of (i, j) Root
According toWithObtain RdisGradient map, be designated asWillMiddle coordinate position is the pixel value note of the pixel of (i, j)
For Calculate againWithMean square error, be designated as Equally, calculateWithMean square error, be designated as
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the picture of (i, j)
The pixel value of vegetarian refreshments,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
The pixel value of point,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j)；
4. distortion model is just perceived using spatial domain, obtains Lorg、Rorg、LdisAnd RdisRespective spatial domain just perceives distortion
Figure, correspondence is designated asWithThen calculateWithMean square error, be designated as Equally, calculateWithMean square error, be designated as
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentIn
Coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of the pixel of (i, j)
Value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j)；
5. light stream matching method is used, I is obtainedorgHorizontal parallax amplitude figure and IorgVertical parallax amplitude figure, correspondence note
ForWithThen basisWithObtain IorgDisparity map, be designated as Dorg, by DorgMiddle coordinate position is (i, j)
The pixel value of pixel is designated as Dorg(i, j),Equally, matched using light stream
Method, obtains IdisHorizontal parallax amplitude figure and IdisVertical parallax amplitude figure, correspondence be designated asWithThen basis
WithObtain IdisDisparity map, be designated as Ddis, by DdisMiddle coordinate position is designated as D for the pixel value of the pixel of (i, j)dis(i,
J),D is calculated afterwardsorgAnd DdisMean square error, be designated as MSEdsp,
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),Represent
Middle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
Pixel value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j)；
6. willAnd MSEdspWhat arranged in sequence was constituted
Set is defined as IdisPerception feature set, be designated as P,
7. use n original undistorted stereo-picture, set up its different type of distortion difference distortion level under
Distortion stereo-picture set, using the distortion stereo-picture set as training set, training set includes several distortion stereo-pictures；So
The mean subjective opinion point of the every width distortion stereo-picture in training set is evaluated using subjective quality assessment method afterwards, will be trained
The mean subjective opinion minute mark of the jth width distortion stereo-picture of concentration is MOSj；According still further to step process 1. to step 6., with
Identical mode obtains the perception feature set of the every width distortion stereo-picture in training set, and the jth width distortion in training set is stood
The perception feature set of body image is designated as Pj；
Wherein, n >=1,1≤j≤S, S represents total width number of the distortion stereo-picture included in training set, MOSj∈[0,
5]；
8. use random forest machine learning algorithm, to training set in the perception feature set of all distortion stereo-pictures enter
Row training so that the regression function value obtained by training is minimum with the error between corresponding mean subjective opinion point, construction
Obtain random forest training pattern；
9. the random forest training pattern for being obtained according to construction, to IdisPerception feature set P tested, prediction is obtained
IdisEvaluating objective quality predicted value, be designated as Qdis, Qdis=MOD (P), wherein, MOD () is the letter of random forest training pattern
Number representation.
Compared with prior art, the advantage of the invention is that：
1) the inventive method weighs the distortion level of stereo-picture with the distortion level of visually-perceptible characteristic pattern, extract with
The related notable figure of viewpoint perceived quality, gradient map and spatial domain just perceive distortion map, extract related to three-dimensional perceived quality
Disparity map, perception feature set is constituted using four kinds of distortion levels of Perception Features figure as the feature of stereo-picture, then with random gloomy
The complicated human visual system of woods machine learning algorithm simulation, carries out the fusion of characteristic parameter, and the inventive method can be objectively
Reflection stereo-picture is subject to the situation of change of visual quality under the influence of various image procossings and compression method, and the inventive method
Evaluation performance do not influenceed by stereoscopic image content and type of distortion, the subjective perception with human eye is consistent.
2) the inventive method is instructed by random forest machine learning algorithm to the perception feature set of distortion stereo-picture
Practice, construction obtains random forest training pattern, it is vertical to distortion to be evaluated further according to the random forest training pattern that construction is obtained
The perception feature set of body image is tested, and prediction obtains the evaluating objective quality predicted value of distortion stereo-picture to be evaluated,
It is this make characteristic parameter with the evaluating objective quality predicted value of optimal amalgamation mode predicted distortion stereo-picture, it is to avoid to people
The correlation properties of class vision system and the complicated simulation process of mechanism, and the perception feature set due to training and the perception of test
Feature set is separate, therefore test result can be avoided to depend on training data unduly such that it is able to effectively carried
Correlation between objective evaluation result high and subjective perception.
Brief description of the drawings
Fig. 1 realizes block diagram for the totality of the inventive method；
Fig. 2 a are the left view dot image of original undistorted horse stereo-pictures；
Fig. 2 b are the distortion left view dot image that the image shown in Fig. 2 a is obtained after JPEG compression；
Fig. 2 c are the notable figure of the image shown in Fig. 2 a；
Fig. 2 d are the notable figure of the image shown in Fig. 2 b；
Fig. 2 e are the gradient map of the image shown in Fig. 2 a；
Fig. 2 f are the gradient map of the image shown in Fig. 2 b；
Fig. 2 g are that the spatial domain of the image shown in Fig. 2 a just perceives distortion map；
Fig. 2 h are that the spatial domain of the image shown in Fig. 2 b just perceives distortion map；
Fig. 2 i are the disparity map of the corresponding stereo-picture of image shown in Fig. 2 a；
Fig. 2 j are the disparity map of the corresponding stereo-picture of image shown in Fig. 2 b；
Fig. 3 a are the left view dot image of Akko (size is 640 × 480) stereo-picture；
Fig. 3 b are the left view dot image of Altmoabit (size is 1024 × 768) stereo-picture；
Fig. 3 c are the left view dot image of Balloons (size is 1024 × 768) stereo-picture；
Fig. 3 d are the left view dot image of Doorflower (size is 1024 × 768) stereo-picture；
Fig. 3 e are the left view dot image of Kendo (size is 1024 × 768) stereo-picture；
Fig. 3 f are the left view dot image of LeaveLaptop (size is 1024 × 768) stereo-picture；
Fig. 3 g are the left view dot image of Lovebierd1 (size is 1024 × 768) stereo-picture；
Fig. 3 h are the left view dot image of Newspaper (size is 1024 × 768) stereo-picture；
Fig. 3 i are the left view dot image of Puppy (size is 720 × 480) stereo-picture；
Fig. 3 j are the left view dot image of Soccer2 (size is 720 × 480) stereo-picture；
Fig. 3 k are the left view dot image of Horse (size is 480 × 270) stereo-picture；
Fig. 3 l are the left view dot image of Xmas (size is 640 × 480) stereo-picture.
Specific embodiment
The present invention is described in further detail below in conjunction with accompanying drawing embodiment.
Human eye is different to the sensitivity of distortion in stereo-picture, therefore the evaluation side of global quality is obtained by local quality
Method is relatively low with the uniformity of subjective perception, and Perception Features figure is shallow-layer reflection of the stereo-picture in human nervous system, is
Stereo-picture is most intuitively experienced in human eye, and when the distortion of stereo-picture causes Perception Features figure distortion, human eye also can be quick
These distortions are discovered on sense ground, therefore, objective evaluation method for quality of stereo images proposed by the present invention extracts four kinds of Perception Features figures,
Distortion map and disparity map are just perceived including notable figure, gradient map, spatial domain, and sense is constituted as characteristic parameter using its distortion level
Know feature set, weigh the distortion level of stereo-picture.Objective evaluation method for quality of stereo images engineering proposed by the present invention
The method of habit carries out Fusion Features, and accuracy highest feature higher-dimension non-thread is obtained by training random forest machine learning algorithm
Property Fusion Model, so as to reach forecasting accuracy higher.
A kind of objective evaluation method for quality of stereo images based on perception feature set proposed by the present invention, its totality realizes frame
Figure is as shown in figure 1, it is comprised the following steps：
1. I is madeorgOriginal undistorted stereo-picture is represented, I is madedisThe stereo-picture of distortion to be evaluated is represented, will
IorgLeft view dot image be designated as Lorg, by IorgRight visual point image be designated as Rorg, by IdisLeft view dot image be designated as Ldis, will
IdisRight visual point image be designated as Rdis。
Fig. 2 a give the left view dot image of original undistorted horse stereo-pictures, and Fig. 2 b are given shown in Fig. 2 a
The distortion left view dot image that is obtained after JPEG compression of image.
2. existing frequency modulation conspicuousness detection algorithm is used, L is obtainedorg、Rorg、LdisAnd RdisRespective notable figure, correspondence is designated asWithThen calculateWithMean square error, be designated as
Equally, calculateWithMean square error, be designated as
Wherein, M represents IorgAnd IdisWidth, N represents IorgAnd IdisHeight, 1≤i≤M, 1≤j≤N,
RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
The pixel value of point,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle seat
Mark is set to the pixel value of the pixel of (i, j).
Fig. 2 c give the notable figure of the image shown in Fig. 2 a, and Fig. 2 d give the notable figure of the image shown in Fig. 2 b.
3. existing horizontal Sobel operators are used, L is obtainedorg、Rorg、LdisAnd RdisRespective horizontal gradient figure, correspondence is designated asWithAnd existing vertical Sobel operators are used, obtain Lorg、Rorg、LdisAnd RdisIt is respective vertical
Gradient map, correspondence is designated asWithThen basisWithObtain LorgGradient map, be designated as
WillMiddle coordinate position is designated as the pixel value of the pixel of (i, j)
And according toWithObtain RorgGradient map, be designated asWillMiddle coordinate position is the pixel of the pixel of (i, j)
Value is designated as According toWithObtain LdisGradient map, be designated as
WillMiddle coordinate position is designated as the pixel value of the pixel of (i, j)
According toWithObtain RdisGradient map, be designated asWillMiddle coordinate position is the pixel value of the pixel of (i, j)
It is designated as Calculate againWithMean square error, be designated as Equally, calculateWithMean square error,
It is designated as
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the picture of (i, j)
The pixel value of vegetarian refreshments,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
The pixel value of point,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j).
Fig. 2 e give the gradient map of the image shown in Fig. 2 a, and Fig. 2 f give the gradient map of the image shown in Fig. 2 b.
4. distortion (Just-Noticeable-Distortion, JND) model is just perceived using existing spatial domain, is obtained
Lorg、Rorg、LdisAnd RdisRespective spatial domain just perceives distortion map, and correspondence is designated asWithThen calculateWithMean square error, be designated as Equally,
CalculateWithMean square error, be designated as
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentIn
Coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of the pixel of (i, j)
Value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j).
Here, by Lorg、Rorg、LdisAnd RdisRespectively as pending image, then the spatial domain for obtaining pending image just may be used
The detailed process for discovering distortion map is：
4. pending image -1, is designated as I；
4. the brightness for -2, obtaining I just perceives distortion map, is designated as JNDl, by JNDlMiddle coordinate position is the pixel of (i, j)
The pixel value of point is designated as JNDl(i, j),Wherein,
Coordinate position is the background value of the pixel of (i, j) in representing I；
4. the texture for -3, obtaining I just perceives distortion map, is designated as JNDt, by JNDtMiddle coordinate position is the pixel of (i, j)
The pixel value of point is designated as JNDt(i, j), JNDt(i, j)=η G (i, j) We(i, j), wherein, η is regulatory factor, and η is taken herein
=0.01, G (i, j) represent I in coordinate position for (i, j) pixel it is maximum under the high-pass filtering operator of different directions
Gradient average value,gradk(i, j) represent I in coordinate position for the pixel of (i, j) exists
Gradient average value under k-th high-pass filtering operator in direction, four direction is horizontally oriented respectively, vertical direction and two it is right
Linea angulata direction, WeCoordinate position is the Weighted Edges factor of the pixel of (i, j) during (i, j) represents I,
4. -4, according to JNDlAnd JNDt, the spatial domain for obtaining I just perceives distortion map, is designated as JNDs, by JNDsMiddle coordinate bit
The pixel value for being set to the pixel of (i, j) is designated as JNDs(i, j), JNDs(i, j)=JNDl(i,j)+JNDt(i,j)-0.3×min
{JNDl(i,j),JNDt(i, j) }, wherein, min () is to take minimum value function.
The spatial domain that Fig. 2 g give the image shown in Fig. 2 a just perceives distortion map, and Fig. 2 h give the image shown in Fig. 2 b
Spatial domain just perceive distortion map.
5. existing light stream matching method is used, I is obtainedorgHorizontal parallax amplitude figure and IorgVertical parallax amplitude figure,
Correspondence is designated asWithThen basisWithObtain IorgDisparity map, be designated as Dorg, by DorgMiddle coordinate position is
The pixel value of the pixel of (i, j) is designated as Dorg(i, j),Equally, use
Existing light stream matching method, obtains IdisHorizontal parallax amplitude figure and IdisVertical parallax amplitude figure, correspondence be designated asWithThen basisWithObtain IdisDisparity map, be designated as Ddis, by DdisMiddle coordinate position is the pixel of (i, j)
Pixel value is designated as Ddis(i, j),D is calculated afterwardsorgAnd DdisMean square error
Difference, is designated as MSEdsp,
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),Represent
Middle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
Pixel value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j).
Fig. 2 i give the disparity map of the corresponding stereo-picture of image shown in Fig. 2 a, and Fig. 2 j give the figure shown in Fig. 2 b
As the disparity map of corresponding stereo-picture.
6. willAnd MSEdspWhat arranged in sequence was constituted
Set is defined as IdisPerception feature set, be designated as P,
7. use n original undistorted stereo-picture, set up its different type of distortion difference distortion level under
Distortion stereo-picture set, using the distortion stereo-picture set as training set, training set includes several distortion stereo-pictures；So
The mean subjective opinion point of the every width distortion stereo-picture in training set is evaluated using existing subjective quality assessment method afterwards,
It is MOS by the mean subjective opinion minute mark of the jth width distortion stereo-picture in training setj；According still further to step 1. to step 6.
Process, obtains the perception feature set of the every width distortion stereo-picture in training set, by the jth width in training set in an identical manner
The perception feature set of distortion stereo-picture is designated as Pj。
Wherein, n >=1, such as takes n=1000,1≤j≤S, and S represents total width of the distortion stereo-picture included in training set
Number, MOSj∈[0,5]。
8. use existing random forest machine learning algorithm, to training set in all distortion stereo-pictures perception it is special
Collection is trained so that the error between by training the regression function value for obtaining to divide with corresponding mean subjective opinion is most
Small, construction obtains random forest training pattern.
9. the random forest training pattern for being obtained according to construction, to IdisPerception feature set P tested, prediction is obtained
IdisEvaluating objective quality predicted value, be designated as Qdis, Qdis=MOD (P), wherein, MOD () is the letter of random forest training pattern
Number representation.
To further illustrate the feasibility and validity of the inventive method, the inventive method is tested.
Random Forest model is made up of 20000 decision trees, and the characteristic for constructing decision tree is 2.
Using 12 undistorted stereo-pictures shown in Fig. 3 a to Fig. 3 l, it is set up different in 5 kinds of different type of distortion
Distortion stereo-picture set under distortion level, including 60 width JPEG compressions coding distortion (JPEG) distortion stereo-picture, 60
The distortion stereogram of the distortion stereo-picture of width JP2000 compressed encodings distortion (JP2K), 60 width white Gaussian noises distortion (WN)
As 312 in the case of, the distortion stereo-picture of 60 width Gaussian Blurs distortion (GB) and 72 width H.264 coding distortion (H.264)
The stereo-picture of width distortion.The respective evaluating objective quality prediction of stereo-picture of the distortion that analysis and utilization the inventive method is obtained
It is worth and the correlation between mean subjective scoring difference.80% for being randomly selected in the stereo-picture of above-mentioned 312 width distortion
The stereo-picture composing training collection of distortion, the stereo-picture of remaining 20% distortion constitutes test set；Then subjective quality is utilized
Evaluation method evaluates the mean subjective opinion point of the stereo-picture of the every width distortion in training set；According still further to step 1. to step
6. process, in an identical manner obtain training set in every width distortion stereo-picture perception feature set；Then use with
Machine forest machine learning algorithm, to training set in the perception feature set of all distortion stereo-pictures be trained so that pass through
The regression function value that training is obtained is minimum with the error between corresponding mean subjective opinion point, and construction obtains random forest training
Model；Afterwards according to the random forest training pattern that obtains of construction, to test set in every width distortion stereo-picture perception
Feature set is tested, and prediction obtains the evaluating objective quality predicted value of the stereo-picture of the every width distortion in test set.
Here, objective parameter as evaluation index, i.e. Pearson is commonly used by the use of assess image quality evaluating method 3
Linear correlation property coefficient (Pearson Linear Correlation Coefficients, PLCC), Spearman rank correlation system
Number (Spearman Rank Order Correlation coefficient, SROCC) and root-mean-square error (Rooted Mean
Squared Error, RMSE).The span of PLCC and SROCC is [0,1], and its value shows that evaluation method is got over closer to 1
It is good, conversely, poorer；RMSE value is smaller, represents that the prediction of evaluation method is more accurate, and performance is better, conversely, then poorer.Expression is commented
PLCC, SROCC and RMSE coefficient of valency performance are as listed in table 1.Knowable to the data listed by table 1, the totality of PLCC and SROCC values
Evaluate more than 0.94, the overall assessment of RMSE is less than 5.8, that is to say, that the solid of the distortion obtained using the inventive method
Correlation between the evaluating objective quality predicted value of image and mean subjective scoring difference is very high, shows objective evaluation knot
Fruit is more consistent with the result of human eye subjective perception, it is sufficient to illustrate the validity of the inventive method.
The evaluating objective quality predicted value and mean subjective of the stereo-picture of the distortion that table 1 is calculated by the inventive method
Correlation between scoring difference
Claims (1)
1. a kind of objective evaluation method for quality of stereo images based on perception feature set, it is characterised in that comprise the following steps：
1. I is madeorgOriginal undistorted stereo-picture is represented, I is madedisThe stereo-picture of distortion to be evaluated is represented, by Iorg's
Left view dot image is designated as Lorg, by IorgRight visual point image be designated as Rorg, by IdisLeft view dot image be designated as Ldis, by IdisThe right side
Visual point image is designated as Rdis；
2. frequency modulation conspicuousness detection algorithm is used, L is obtainedorg、Rorg、LdisAnd RdisRespective notable figure, correspondence is designated asWithThen calculateWithMean square error, be designated as
Equally, calculateWithMean square error, be designated as
Wherein, M represents IorgAnd IdisWidth, N represents IorgAnd IdisHeight, 1≤i≤M, 1≤j≤N,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
Pixel value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate bit
It is set to the pixel value of the pixel of (i, j)；
3. horizontal Sobel operators are used, L is obtainedorg、Rorg、LdisAnd RdisRespective horizontal gradient figure, correspondence is designated asWithAnd vertical Sobel operators are used, obtain Lorg、Rorg、LdisAnd RdisRespective vertical gradient map,
Correspondence is designated asWithThen basisWithObtain LorgGradient map, be designated asWill
Middle coordinate position is designated as the pixel value of the pixel of (i, j) And root
According toWithObtain RorgGradient map, be designated asWillMiddle coordinate position is the pixel value note of the pixel of (i, j)
For According toWithObtain LdisGradient map, be designated asWillMiddle coordinate position is designated as the pixel value of the pixel of (i, j)
According toWithObtain RdisGradient map, be designated asWillMiddle coordinate position is the pixel value of the pixel of (i, j)
It is designated as Calculate againWithMean square error, be designated as Equally, calculateWithMean square error,
It is designated as
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),Represent
Middle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
Pixel value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),Represent
Middle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of (i, j)
Pixel value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentIn
Coordinate position is the pixel value of the pixel of (i, j)；
4. distortion model is just perceived using spatial domain, according to Lorg、Rorg、LdisAnd RdisRespective brightness just perceives distortion map and line
The just discernable distortion map of reason obtains corresponding spatial domain and just perceives distortion map respectively, and correspondence is designated as WithSo
After calculateWithMean square error, be designated as Together
Sample, calculatesWithMean square error, be designated as
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate
Position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel value of the pixel of (i, j)；
5. light stream matching method is used, I is obtainedorgHorizontal parallax amplitude figure and IorgVertical parallax amplitude figure, correspondence be designated as
WithThen basisWithObtain IorgDisparity map, be designated as Dorg, by DorgMiddle coordinate position is the pixel of (i, j)
The pixel value of point is designated as Dorg(i, j),Equally, using light stream matching method,
Obtain IdisHorizontal parallax amplitude figure and IdisVertical parallax amplitude figure, correspondence be designated asWithThen basisWithObtain IdisDisparity map, be designated as Ddis, by DdisMiddle coordinate position is designated as D for the pixel value of the pixel of (i, j)dis(i,
J),D is calculated afterwardsorgAnd DdisMean square error, be designated as MSEdsp,
Wherein,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j),RepresentMiddle seat
Mark is set to the pixel value of the pixel of (i, j),RepresentMiddle coordinate position is the pixel of the pixel of (i, j)
Value,RepresentMiddle coordinate position is the pixel value of the pixel of (i, j)；
6. willAnd MSEdspThe set that arranged in sequence is constituted
It is defined as IdisPerception feature set, be designated as P,
7. n original undistorted stereo-picture is used, its distortion under different type of distortion difference distortion level is set up
Stereo-picture set, using the distortion stereo-picture set as training set, training set includes several distortion stereo-pictures；Then it is sharp
The mean subjective opinion point of the every width distortion stereo-picture in training set is evaluated with subjective quality assessment method, by training set
Jth width distortion stereo-picture mean subjective opinion minute mark be MOSj；According still further to step process 1. to step 6., with identical
Mode obtain the perception feature set of the every width distortion stereo-picture in training set, by the jth width distortion stereogram in training set
The perception feature set of picture is designated as Pj；
Wherein, n >=1,1≤j≤S, S represents total width number of the distortion stereo-picture included in training set, MOSj∈[0,5]；
8. use random forest machine learning algorithm, to training set in the perception feature set of all distortion stereo-pictures instruct
Practice so that the regression function value obtained by training is minimum with the error between corresponding mean subjective opinion point, and construction is obtained
Random forest training pattern；
9. the random forest training pattern for being obtained according to construction, to IdisPerception feature set P tested, prediction obtain Idis's
Evaluating objective quality predicted value, is designated as Qdis, Qdis=MOD (P), wherein, MOD () is the function table of random forest training pattern
Show form.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510303868.3A CN104954778B (en) | 2015-06-04 | 2015-06-04 | Objective stereo image quality assessment method based on perception feature set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510303868.3A CN104954778B (en) | 2015-06-04 | 2015-06-04 | Objective stereo image quality assessment method based on perception feature set |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104954778A CN104954778A (en) | 2015-09-30 |
| CN104954778B true CN104954778B (en) | 2017-05-24 |
Family
ID=54169076
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510303868.3A Active CN104954778B (en) | 2015-06-04 | 2015-06-04 | Objective stereo image quality assessment method based on perception feature set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104954778B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105205504B (en) * | 2015-10-04 | 2018-09-18 | 北京航空航天大学 | A kind of image attention regional quality evaluation index learning method based on data-driven |
| CN105376563B (en) * | 2015-11-17 | 2017-03-22 | 浙江科技学院 | No-reference three-dimensional image quality evaluation method based on binocular fusion feature similarity |
| CN105611272B (en) * | 2015-12-28 | 2017-05-03 | 宁波大学 | Eye exactly perceptible stereo image distortion analyzing method based on texture complexity |
| CN106412569B (en) * | 2016-09-28 | 2017-12-15 | 宁波大学 | A kind of selection of feature based without referring to more distortion stereo image quality evaluation methods |
| CN106780476A (en) * | 2016-12-29 | 2017-05-31 | 杭州电子科技大学 | A kind of stereo-picture conspicuousness detection method based on human-eye stereoscopic vision characteristic |
| CN106803952B (en) * | 2017-01-20 | 2018-09-14 | 宁波大学 | In conjunction with the cross validation depth map quality evaluating method of JND model |
| CN108401150B (en) * | 2018-03-22 | 2019-08-27 | 浙江科技学院 | A kind of compressed sensing reconstruction algorithm statistic of attribute evaluation method of analog vision subjective perception |
| CN109872305B (en) * | 2019-01-22 | 2020-08-18 | 浙江科技学院 | No-reference stereo image quality evaluation method based on quality map generation network |
| CN114567776B (en) * | 2022-02-21 | 2023-05-05 | 宁波职业技术学院 | Video low-complexity coding method based on panoramic visual perception characteristics |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103200420B (en) * | 2013-03-19 | 2015-03-25 | 宁波大学 | Three-dimensional picture quality objective evaluation method based on three-dimensional visual attention |
| KR101550065B1 (en) * | 2013-06-21 | 2015-09-07 | 한국과학기술원 | Stereoscopic imaging method and system for visually comfortable 3D images |
| CN103581661B (en) * | 2013-10-28 | 2015-06-03 | 宁波大学 | Method for evaluating visual comfort degree of three-dimensional image |
| CN104394403B (en) * | 2014-11-04 | 2016-04-27 | 宁波大学 | A kind of stereoscopic video quality method for objectively evaluating towards compression artefacts |
-
2015
- 2015-06-04 CN CN201510303868.3A patent/CN104954778B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN104954778A (en) | 2015-09-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104954778B (en) | Objective stereo image quality assessment method based on perception feature set | |
| CN102209257B (en) | Stereo image quality objective evaluation method | |
| CN103581661B (en) | Method for evaluating visual comfort degree of three-dimensional image | |
| CN105407349B (en) | No-reference objective three-dimensional image quality evaluation method based on binocular visual perception | |
| CN102333233A (en) | Stereo image quality objective evaluation method based on visual perception | |
| CN107635136B (en) | View-based access control model perception and binocular competition are without reference stereo image quality evaluation method | |
| CN103426173B (en) | Objective evaluation method for stereo image quality | |
| CN104811691B (en) | A kind of stereoscopic video quality method for objectively evaluating based on wavelet transformation | |
| CN102708567B (en) | Visual perception-based three-dimensional image quality objective evaluation method | |
| CN104902268B (en) | Based on local tertiary mode without with reference to three-dimensional image objective quality evaluation method | |
| CN105338343A (en) | No-reference stereo image quality evaluation method based on binocular perception | |
| CN109242834A (en) | It is a kind of based on convolutional neural networks without reference stereo image quality evaluation method | |
| Geng et al. | A stereoscopic image quality assessment model based on independent component analysis and binocular fusion property | |
| CN102547368A (en) | Objective evaluation method for quality of stereo images | |
| CN107330873B (en) | Stereo image quality objective evaluation method based on multi-scale binocular fusion and local feature extraction | |
| CN103873854A (en) | Method for determining number of stereoscopic image subjective assessment testees and experiment data | |
| CN109345502A (en) | A kind of stereo image quality evaluation method based on disparity map stereochemical structure information extraction | |
| CN104361583B (en) | A kind of method determining asymmetric distortion three-dimensional image objective quality | |
| CN102708568A (en) | Stereoscopic image objective quality evaluation method on basis of structural distortion | |
| CN103841411B (en) | A kind of stereo image quality evaluation method based on binocular information processing | |
| CN105898279B (en) | A kind of objective evaluation method for quality of stereo images | |
| CN105488792B (en) | Based on dictionary learning and machine learning without referring to stereo image quality evaluation method | |
| CN103108209B (en) | Stereo image objective quality evaluation method based on integration of visual threshold value and passage | |
| CN102999911B (en) | Three-dimensional image quality objective evaluation method based on energy diagrams | |
| CN107018410A (en) | A kind of non-reference picture quality appraisement method based on pre- attention mechanism and spatial dependence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |