CN108257131A - A kind of 3D rendering quality evaluating method - Google Patents
A kind of 3D rendering quality evaluating method Download PDFInfo
- Publication number
- CN108257131A CN108257131A CN201810158525.6A CN201810158525A CN108257131A CN 108257131 A CN108257131 A CN 108257131A CN 201810158525 A CN201810158525 A CN 201810158525A CN 108257131 A CN108257131 A CN 108257131A
- Authority
- CN
- China
- Prior art keywords
- distortion
- rendering
- represent
- quality
- layer
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20228—Disparity calculation for image-based rendering
Abstract
The present invention proposes a kind of 3D rendering quality evaluating method, includes the following steps：Step 1）By the type of distortion that the natural scene feature of 2D images, prediction 3D rendering or so view is extracted in wavelet field；The energy spectrum statistical nature of disparity map is extracted, predicts the type of distortion of 3D rendering disparity map；Classified by depth belief network to the type of distortion of 3D rendering；Step 2）The mapping relations established between statistical nature and the quality of image obtain the quality of 3D rendering.Advantageous effect：The algorithm that technical scheme is proposed is compared with the algorithm of existing 3D rendering, the consistency higher of subjective quality assessment, and universal more preferable.
Description
Technical field
The invention belongs to technical field of computer vision more particularly to a kind of 3D rendering quality evaluating methods.
Background technology
Vision is one of important channel that people obtain external information, with three-dimensional movie and the rapid hair of display equipment
Exhibition, 3 D video and image are becoming one of multimedia form most popular in our daily lifes.Due to bandwidth and set
The limitation of standby physical characteristic during the acquisition, transmission and display of 3-D view, can lose by different type and in various degree
(Zhang Yan, Anping, Zhang Qiuwen wait binocular tri-dimensional video minimum discernable distortion models and its in quality evaluation for the influence of proper class type
Application [J] electronics and information journal, 2012,34 (3):698-703), the 3D videos and picture quality that people obtain are often not
The needs of people can be reached, it is therefore desirable to consider the three-dimensional stereoscopic three-dimensional image quality evaluation index perceived.Three-dimensional image quality
Evaluation (IQA) is the important subject of image processing field.It can be used for real-time system detection image quality, can also be used as weighing apparatus
Measure the standard of 3-D view Processing Algorithm.Compared with traditional two dimensional image, 3-D view is made of 3 parts：Left and right view
And depth information.Therefore, the quality evaluation of three-dimensional image is more much more complex than two dimensional image.
According to the depth information that 3D rendering whether is considered during 3D rendering quality evaluation, we are by existing 3D
Image quality evaluation algorithm is divided into two classes (Shao F, Lin W, Gu S, et al.Perceptual full-reference
quality assessment of stereoscopic images by considering binocular visual
characteristics[J].IEEE Transactions on Image Processing,2013,22(5):1940-
1953.).The first kind is the index of two dimension measurement, i.e., using traditional 2D image quality evaluation indexs, such as document (Song Y,
Yu M,Zheng K H,et al.New Objective Stereo Image Quality Metric Using Human
Visual Characteristics and Phase Congruency[J].Advanced Materials Research,
2013,816:506-511.), it can be good at predicting the quality of 2D images, a left side for 3D rendering obtained using this one kind 2D index
The quality of right view finally obtains the quality of 3D rendering, and algorithm is relatively easy, but due to having ignored the depth information of 3D rendering,
Algorithm and human subject's mass are inconsistent.Second class algorithm considers the distinctive depth information of 3D rendering.Since binocular parallax can
Depth information is provided, depth information is considered as an important factor for related with 3D vision perception, and to X-Y scheme by the second class measurement
The quality and depth information of picture are assessed.The score of two dimensional image with depth quality is combined, forms obtaining for 3-D view
Point.Zhang(Zhang L,Tong M H,Marks T K,et al.SUN:A Bayesian framework for
saliency using natural statistics[J].Journal of Vision.2008,8(7):1-20.) et al. adopt
The quality of image is calculated with two dimension blunt C4 and structural similarity SSIM, the quality of 3D rendering is obtained with reference to two kinds of quality.You
(You J,Korhonen J,Perkis A.Spatial and temporal pooling of image quality
metrics for perceptual video quality assessment on packet loss streams[C]
.2010IEEE International Conference on Acoustics Speech and Signal Processing
(ICASSP),2010:It 1002-1005.) et al. attempts many two dimension measurements and applies to 3-D view, it is proposed that a two dimension regards
The SSIM scoring binding models of figure and the antipode of average disparity map obtain 3D rendering quality.However these evaluation indexes for
The 3D rendering of symmetrical distortion can obtain good effect, but then unsatisfactory for the 3D rendering of asymmetric distortion.In order to
Symmetrical distortion and asymmetric distortion are improved in subjective consistency, this paper presents a kind of commenting based on 3D rendering type of distortion
Valency algorithm.Algorithm is first according to the type of distortion of the natural scene statistical nature prognostic chart picture of 3D rendering, symmetrical distortion and non-right
Claim distortion, then set up the mapping relations between each distortion and picture quality, the quality score being distorted with reference to two classes obtains
The quality of image.
Invention content
The purpose of the present invention is to overcome the deficiency in the prior art, provide it is a kind of based on natural scene statistical nature without ginseng
3D rendering quality evaluation algorithm is examined, is specifically realized by following technical scheme：
The 3D rendering quality evaluating method, includes the following steps：
The distortion class that step 1) passes through the natural scene feature in wavelet field extraction 2D images, prediction 3D rendering or so view
Type；The energy spectrum statistical nature of disparity map is extracted, predicts the type of distortion of 3D rendering disparity map；By depth belief network to 3D
Image is classified；
The mapping relations that step 2) is established between statistical nature and the quality of image obtain the quality of 3D rendering.
The further design of the 3D rendering quality evaluating method is that left and right view, which is distorted, in step 3) includes five kinds of mistakes
Proper class type, five kinds of type of distortion are respectively WN, FastFading, JPEG, JPEG2000 and Blur；Disparity map distortion is by carrying
The energy spectrum statistical nature of disparity map is taken to carry out predicted distortion type, including two kinds of type of distortion, two kinds of type of distortion are respectively
Symmetrical distortion and asymmetric distortion.
The further design of the 3D rendering quality evaluating method is that the step 3) obtains 3D rendering according to formula (1)
Quality,
Wherein piRepresent probability, the p of symmetrical distortionijRepresent probability, the q of asymmetric distortionijRepresent five kinds of distortions
Probability, QL, QRRepresent the quality of left and right view.
The further design of the 3D rendering quality evaluating method is that depth belief network includes one in the step 1)
A visual layers, three hidden layers and a recurrence layer, the input of visual layers is the NSS feature vectors of 3D rendering or so view,
Each visual layers includes 220 nodes；Hidden layer includes 100 nodes；Linear regression layer includes two nodes, represents respectively
Symmetrical distortion and asymmetric distortion；Visual layers and the joint probability distribution of hidden layer are obtained by formula (2).
p(X,h1,h2,h3)=p (X | h1)·p(h1|h2)·p(h2,h3) (2)
In formula, X represents input left image feature or right image feature, h1、h2、h3Accordingly represent three hidden layers.
The further design of the 3D rendering quality evaluating method is, needed in the step 1) to depth belief network into
Row adjustment, adjustment mode are as follows：
1) unsupervised learning mechanism：Each layer of hidden layer trains each layer weights by greedy learning method, from bottom to
On successively training obtain training result, first layer is set as second order Gauss model, and every layer is relatively independent；
2) strategy of adjustment supervision in real time：It is adjusted in real time according to weighted value of the training result to each layer；
3) linear regression：It is supervised by unsupervised learning and adjustment, obtains each layer of weight, then set up NSS features
Output and type of distortion between regression model.
The further design of the 3D rendering quality evaluating method is that the acquisition of energy spectrum statistical nature includes following step
Suddenly：
A) it is defined as follows formula (3)；
CI (x, y)=WL(x,y)·IL(x,y)+WR(x+d,y)·IR(x+d, y) (3)
IL、IR、WLAnd WRThe parallax value of left and right view is represented respectively, and d represents weighted value；
B) W is obtained according to formula (4)LAnd WR,
Wherein,GEL(x, y) andGERIt is the energy response in all scales and direction that (x+d, y), which represents left and right view,
(x, y) represents the position of wave filter；
C) gradient information of image is obtained by formula (5),
In formula (5), Gx(x, y) represents horizontal direction Grad, Gy(x, y) represents that vertical gradient value, G (x, y) represent
Gradient magnitude and α (x, y) represent gradient direction；
D) finally, the monocular feature of stereo-picture and binocular feature are cascaded, obtains the left and right image f of imageL,
fR,fL=[gL, c] and fR=[gR,c]。
The further design of the 3D rendering quality evaluating method is that the natural scene feature of 2D images passes through formula
(6) it obtains,
In formula (6), W, H represent width and height respectively, and x takes the positive integer of 1,2...W, and y takes the positive integer of 1,2...H,
σθ,λ(x, y) represents standard deviation, Sθ,λ(x, y) represents divergence, GEθ,λ(x, y) represent by Gabor filter response amplitude,
μθ,λ(x, y) represents that the mean value of response amplitude, λ represent that wavelength, θ represent centric angle, controls the filtering direction of wave filter.
Advantages of the present invention is as follows：
3D rendering type of distortion is divided into symmetrical and asymmetric distortion, Jin Erjian by the 3D rendering quality evaluating method of the present invention
Mapping of the vertical natural scene feature (Natural Scene Statistics, hereinafter NSS features) between picture quality
Relationship obtains the quality of 3D rendering, the algorithm of the algorithm that the results show technical scheme is proposed and existing 3D rendering
It compares, the consistency higher of subjective quality assessment, and universal more preferable.
Description of the drawings
Fig. 1 is 3D NR model frameworks of the present invention.
Fig. 2 is Deep Belief Networks (DBNs) structure diagram of the present invention.
Fig. 3 is the accuracy comparison schematic diagram of type of distortion of the present invention classification.
Fig. 4 is the dependency diagram between subjective quality.
Specific embodiment
Technical scheme of the present invention is further illustrated with attached drawing in conjunction with specific embodiments.
Such as Fig. 1,3D rendering quality evaluating method provided in this embodiment, including descending step as follows：
The distortion class that step 1) passes through the natural scene feature in wavelet field extraction 2D images, prediction 3D rendering or so view
Type；Extract the energy spectrum statistical nature of disparity map.
The mapping relations that step 2) is established between statistical nature and the quality of image obtain the quality of 3D rendering.
View distortion in left and right includes five kinds of type of distortion in step 3), five kinds of type of distortion be respectively WN, FastFading,
JPEG, JPEG2000 and Blur；Disparity map distortion carries out predicted distortion class by extracting the energy spectrum statistical nature of disparity map
Type, including two kinds of type of distortion, two kinds of type of distortion are respectively symmetrical distortion and asymmetric distortion.
The further design of the 3D rendering quality evaluating method is that the step 3) obtains 3D rendering according to formula (1)
Quality,
Wherein piRepresent probability, the p of symmetrical distortionijRepresent probability, the q of asymmetric distortionijRepresent above-mentioned common five kinds of mistakes
Genuine probability, QL, QRRepresent the quality of left and right view.
In the step 1)
p(X,h1,h2,h3)=p (X | h1)·p(h1|h2)·p(h2,h3) (2)
In formula, X represents the left image feature of input or right image feature, h1、h2、h3Accordingly represent three hidden layers.
In order to improve the classification accuracy of this paper algorithms, depth belief network (Deep Belief are used in step 1)
Nets, hereinafter DBNs) classify to disparity map type of distortion.The results show, the classification accuracy of this paper algorithms
Mean value is up to 86.6%, standard deviation 2.88.Specific classification results such as Fig. 3, Fig. 3 show DBNs graders employed herein
Compared to svm classifier accuracy higher.The DBNs of the present embodiment includes a visual layers, three hidden layers and a recurrence
Layer, the input of visual layers is the NSS feature vectors of 3D rendering or so view, each visual layers includes 220 nodes；Hidden layer
Include 100 nodes；Linear regression layer includes two nodes, represents symmetrical distortion and asymmetric distortion respectively；Visual layers and hidden
The joint probability distribution for hiding layer is obtained by formula (2).
It needs to be adjusted DBNs in step 1), adjustment mode is as follows：
1) unsupervised learning mechanism：Each layer can regard RBM (limited Boltzmann machine) as, pass through greedy study side
Method trains each layer weights, is successively trained from bottom-up, first layer is set as second order Gauss model, and every layer relatively independent；
2) strategy of adjustment supervision in real time：It is adjusted in real time according to weighted value of the prediction result to each layer；
3) linear regression：By unsupervised training and the fine tuning of supervision, each layer of weight is obtained, then sets up NSS spies
Regression model between the output of sign and type of distortion.
The acquisition of 3D natural scene features includes the following steps：
A) it is defined as follows formula (3)；
CI (x, y)=WL(x,y)·IL(x,y)+WR(x+d,y)·IR(x+d, y) (3)
IL、IR、WLAnd WRThe parallax value of left and right view is represented respectively, and d represents weighted value；
B) W is obtained according to formula (4)LAnd WR,
Wherein,GEL(x, y) andGERIt is the energy response in all scales and direction that (x+d, y), which represents left and right view,
(x, y) represents the position of wave filter.
C) gradient information of image is obtained by formula (5),
In formula (5), Gx(x, y) represents horizontal direction Grad, Gy(x, y) represents that vertical gradient value, G (x, y) represent
Gradient magnitude and α (x, y) represent gradient direction；
D) finally, the monocular feature of stereo-picture and binocular feature are cascaded, obtains the left and right image f of imageL,
fR,fL=[gL, c] and fR=[gR,c]。
The further design of the 3D rendering quality evaluating method is that the natural scene feature of 2D images passes through formula
(6) it obtains.
In formula (6), W, H represent width and height respectively, and x takes the positive integer of 1,2...W, and y takes the positive integer of 1,2...H,
σθ,λ(x, y) represents standard deviation, Sθ,λ(x, y) represents divergence, GEθ,λ(x, y) represent by Gabor filter response amplitude,
μθ,λ(x, y) represents that the mean value of response amplitude, λ represent that wavelength, θ represent centric angle, controls the filtering direction of wave filter.
The present embodiment is using Spearman rank correlation coefficient (SROCC) and Pearson's linearly dependent coefficient (LCC) to prediction
As a result it measures.The value of SROCC and LCC is higher, and the performance for showing algorithm is better.Simultaneously by this paper algorithms with it is existing common
3D non-reference picture quality appraisement algorithms are compared.
Table I and Table II are shown in I in the database and II, this paper algorithm and BRISQUE, You, Hewage, MS-SSIM are calculated
Method is in SROCC and compares, and experimental result shows the 3D rendering prediction for symmetrical distortion, and this paper algorithms are superior to existing 3D figures
As quality evaluation algorithm, Table II shows that, corresponding to the comparison carried out in asymmetric distortion library, comparison result is shown for asymmetric mistake
For the quality evaluation of true image, this paper algorithms also have certain advantage.And due to not needing to any reference information, herein
Algorithm has widely application.
Table I common algorithms and this paper algorithms are in database I compared with SROCC
Table II common algorithms and this paper algorithms are in database II compared with SROCC
| Algorithms | WN | JP2K | JPEG | Blur | FF | Combine |
| BRISQUE | 0.932 | 0.822 | 0.560 | 0.720 | 0.840 | 0.783 |
| You | 0.909 | 0.894 | 0.795 | 0.813 | 0.891 | 0786 |
| MS-SSIM | 0.980 | 0.841 | 0.842 | 0.908 | 0.884 | 0.889 |
| Hewage | 0.880 | 0.598 | 0.736 | 0.028 | 0.684 | 0.501 |
| Our Method | 0.951 | 0.870 | 0.850 | 0.912 | 0.942 | 0.899 |
Table III and Table IV show this paper algorithms on LCC with BRISQUE, You, Hewage, MS-SSIM algorithm comparisons are real
It tests result and shows that this paper algorithms are better than other several algorithms.
Fig. 4 shows the fitness between this paper algorithms and subjective perception, as seen from the figure, proposed algorithm and subjectivity
It is similar to linear relationship between quality evaluation, meets human visual perception system.
Table III common algorithms and this paper algorithms are in database I compared with LCC
| Algorithms | WN | JP2K | JPEG | Blur | FF | Combine |
| BRISQUE | 0.941 | 0.847 | 0.615 | 0.926 | 0.852 | 0.912 |
| You | 0.941 | 0.877 | 0.487 | 0.919 | 0.930 | 0.881 |
| MS-SSIM | 0.942 | 0.912 | 0.603 | 0.942 | 0.776 | 0.917 |
| Hewage | 0.895 | 0.904 | 0.530 | 0.798 | 0.669 | 0.830 |
| Our Method | 0.876 | 0.923 | 0.693 | 0.820 | 0.855 | 0.912 |
Table IV common algorithms and this paper algorithms are in database II compared with LCC
| Algorithm | WN | JP2K | JPEG | Blur | FF | Combine |
| BRISQUE | 0.823 | 0.840 | 0.650 | 0.936 | 0.870 | 0.892 |
| You | 0.912 | 0.905 | 0.830 | 0.784 | 0.915 | 0.800 |
| MS-SSIM | 0.957 | 0.834 | 0.862 | 0.963 | 0.901 | 0.900 |
| Hewage | 0.891 | 0.662 | 0.734 | 0.450 | 0.745 | 0.60 |
| Our Method | 0.950 | 0.91 | 0.941 | 0.94 | 0.912 | 0.912 |
The foregoing is only a preferred embodiment of the present invention, but protection scope of the present invention be not limited thereto,
Any one skilled in the art in the technical scope disclosed by the present invention, the change or replacement that can be readily occurred in,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of the claims
Subject to.
Claims (7)
1. a kind of 3D rendering quality evaluating method, it is characterised in that include the following steps：
The type of distortion that step 1) passes through the natural scene feature in wavelet field extraction 2D images, prediction 3D rendering or so view；
The energy spectrum statistical nature of disparity map is extracted, predicts the type of distortion of 3D rendering disparity map；3D is schemed by depth belief network
As classifying；
The mapping relations that step 2) is established between statistical nature and the quality of image obtain the quality of 3D rendering.
2. 3D rendering quality evaluating method according to claim 1, it is characterised in that left and right view distortion packet in step 3)
Five kinds of type of distortion are included, five kinds of type of distortion are respectively WN, FastFading, JPEG, JPEG2000 and Blur；Disparity map loses
Predicted distortion type very is carried out by extracting the energy spectrum statistical nature of disparity map, including two kinds of type of distortion, two kinds of distortion classes
Type is respectively symmetrical distortion and asymmetric distortion.
3. 3D rendering quality evaluating method according to claim 2, it is characterised in that the step 3) is obtained according to formula (1)
The quality of 3D rendering,
Wherein piRepresent probability, the p of symmetrical distortionijRepresent probability, the q of asymmetric distortionijRepresent the probability of five kinds of distortions,
QL, QRRepresent the quality of left and right view.
4. 3D rendering quality evaluating method according to claim 1, it is characterised in that depth conviction net in the step 1)
Network includes a visual layers, three hidden layers and a recurrence layer, and the input of visual layers is the NSS spies of 3D rendering or so view
Sign vector, each visual layers include 220 nodes；Hidden layer includes 100 nodes；Linear regression layer includes two nodes,
Symmetrical distortion and asymmetric distortion are represented respectively；Visual layers and the joint probability distribution of hidden layer are obtained by formula (2).
p(X,h1,h2,h3)=p (X | h1)·p(h1|h2)·p(h2,h3) (2)
In formula, X represents input left image feature or right image feature, h1、h2、h3Accordingly represent three hidden layers.
5. 3D rendering quality evaluating method according to claim 1, it is characterised in that need to believe depth in the step 1)
It reads network to be adjusted, adjustment mode is as follows：
1) unsupervised learning mechanism：Each layer of hidden layer trains each layer weights by greedy learning method, from it is bottom-up by
Layer training obtains training result, and first layer is set as second order Gauss model, and every layer relatively independent；
2) strategy of adjustment supervision in real time：It is adjusted in real time according to weighted value of the training result to each layer；
3) linear regression：It is supervised by unsupervised learning and adjustment, obtains each layer of weight, then set up the defeated of NSS features
Go out the regression model between type of distortion.
6. 3D rendering quality evaluating method according to claim 1, it is characterised in that the acquisition packet of energy spectrum statistical nature
Include following steps：
A) it is defined as follows formula (3)；
CI (x, y)=WL(x,y)·IL(x,y)+WR(x+d,y)·IR(x+d, y) (3)
IL、IR、WLAnd WRThe parallax value of left and right view is represented respectively, and d represents weighted value；
B) W is obtained according to formula (4)LAnd WR,
Wherein, GEL(x, y) and GERIt is the energy response in all scales and direction that (x+d, y), which represents left and right view, (x,
Y) position of wave filter is represented；
C) gradient information of image is obtained by formula (5),
In formula (5), Gx(x, y) represents horizontal direction Grad, Gy(x, y) represents that vertical gradient value, G (x, y) represent gradient
Amplitude and α (x, y) represent gradient direction；
D) finally, the monocular feature of stereo-picture and binocular feature are cascaded, obtains the left and right image f of imageL,fR,fL
=[gL, c] and fR=[gR,c]。
7. 3D rendering quality evaluating method according to claim 1, it is characterised in that the natural scene feature of 2D images is led to
Formula (6) acquisition is crossed,
In formula (6), W, H represent width and height respectively, and x takes the positive integer of 1,2...W, and y takes the positive integer of 1,2...H, σθ,λ
(x, y) represents standard deviation, Sθ,λ(x, y) represents divergence, GEθ,λ(x, y) represents the amplitude responded by Gabor filter, μθ,λ
(x, y) represents that the mean value of response amplitude, λ represent that wavelength, θ represent centric angle, controls the filtering direction of wave filter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810158525.6A CN108257131A (en) | 2018-02-24 | 2018-02-24 | A kind of 3D rendering quality evaluating method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810158525.6A CN108257131A (en) | 2018-02-24 | 2018-02-24 | A kind of 3D rendering quality evaluating method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108257131A true CN108257131A (en) | 2018-07-06 |
Family
ID=62745282
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810158525.6A Pending CN108257131A (en) | 2018-02-24 | 2018-02-24 | A kind of 3D rendering quality evaluating method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108257131A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945552A (en) * | 2012-10-22 | 2013-02-27 | 西安电子科技大学 | No-reference image quality evaluation method based on sparse representation in natural scene statistics |
| CN106960432A (en) * | 2017-02-08 | 2017-07-18 | 宁波大学 | One kind is without with reference to stereo image quality evaluation method |
| CN107635136A (en) * | 2017-09-27 | 2018-01-26 | 北京理工大学 | View-based access control model is perceived with binocular competition without with reference to stereo image quality evaluation method |
-
2018
- 2018-02-24 CN CN201810158525.6A patent/CN108257131A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945552A (en) * | 2012-10-22 | 2013-02-27 | 西安电子科技大学 | No-reference image quality evaluation method based on sparse representation in natural scene statistics |
| CN106960432A (en) * | 2017-02-08 | 2017-07-18 | 宁波大学 | One kind is without with reference to stereo image quality evaluation method |
| CN107635136A (en) * | 2017-09-27 | 2018-01-26 | 北京理工大学 | View-based access control model is perceived with binocular competition without with reference to stereo image quality evaluation method |
Non-Patent Citations (2)
| Title |
|---|
| YANQING LI ET.AL: "No-Reference Stereoscopic Image Quality Assessment Using Natural Scene Statistics", 《2017 2ND INTERNATIONAL CONFERENCE ON MULTIMEDIA AND IMAGE PROCESSING》 * |
| 田维军 等: "基于深度学习的无参考立体图像质量评价", 《计算机辅助设计与图形学学报》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Learning structure of stereoscopic image for no-reference quality assessment with convolutional neural network | |
| Fang et al. | Objective quality assessment of screen content images by uncertainty weighting | |
| Chao et al. | Salgan360: Visual saliency prediction on 360 degree images with generative adversarial networks | |
| CN105744256B (en) | Based on the significant objective evaluation method for quality of stereo images of collection of illustrative plates vision | |
| CN108765414B (en) | No-reference stereo image quality evaluation method based on wavelet decomposition and natural scene statistics | |
| CN104537647B (en) | A kind of object detection method and device | |
| CN108391121B (en) | No-reference stereo image quality evaluation method based on deep neural network | |
| CN109255358B (en) | 3D image quality evaluation method based on visual saliency and depth map | |
| CN107396095B (en) | A kind of no reference three-dimensional image quality evaluation method | |
| CN104994375A (en) | Three-dimensional image quality objective evaluation method based on three-dimensional visual saliency | |
| CN104867138A (en) | Principal component analysis (PCA) and genetic algorithm (GA)-extreme learning machine (ELM)-based three-dimensional image quality objective evaluation method | |
| CN109831664B (en) | Rapid compressed stereo video quality evaluation method based on deep learning | |
| CN107481236A (en) | A kind of quality evaluating method of screen picture | |
| CN107959848A (en) | Universal no-reference video quality evaluation algorithms based on Three dimensional convolution neutral net | |
| CN109242834A (en) | It is a kind of based on convolutional neural networks without reference stereo image quality evaluation method | |
| CN110674925B (en) | No-reference VR video quality evaluation method based on 3D convolutional neural network | |
| CN105654142A (en) | Natural scene statistics-based non-reference stereo image quality evaluation method | |
| CN108259893B (en) | Virtual reality video quality evaluation method based on double-current convolutional neural network | |
| CN109685772B (en) | No-reference stereo image quality evaluation method based on registration distortion representation | |
| CN104866864A (en) | Extreme learning machine for three-dimensional image quality objective evaluation | |
| Tan et al. | Computational aesthetics of photos quality assessment based on improved artificial neural network combined with an autoencoder technique | |
| CN111915589A (en) | Stereo image quality evaluation method based on hole convolution | |
| CN105898279B (en) | A kind of objective evaluation method for quality of stereo images | |
| CN109523590B (en) | 3D image depth information visual comfort evaluation method based on sample | |
| CN108848365B (en) | A kind of reorientation stereo image quality evaluation method |
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: 20180706 |