CN102708568A - Stereoscopic image objective quality evaluation method on basis of structural distortion - Google Patents

Stereoscopic image objective quality evaluation method on basis of structural distortion Download PDF

Info

Publication number
CN102708568A
CN102708568A CN2012101450340A CN201210145034A CN102708568A CN 102708568 A CN102708568 A CN 102708568A CN 2012101450340 A CN2012101450340 A CN 2012101450340A CN 201210145034 A CN201210145034 A CN 201210145034A CN 102708568 A CN102708568 A CN 102708568A
Authority
CN
China
Prior art keywords
coordinate position
matrix
referred
represents
distortion
Prior art date
Application number
CN2012101450340A
Other languages
Chinese (zh)
Other versions
CN102708568B (en
Inventor
周俊明
彭宗举
毛香英
王晓东
蒋刚毅
邵枫
郁梅
Original Assignee
宁波大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 宁波大学 filed Critical 宁波大学
Priority to CN 201210145034 priority Critical patent/CN102708568B/en
Publication of CN102708568A publication Critical patent/CN102708568A/en
Application granted granted Critical
Publication of CN102708568B publication Critical patent/CN102708568B/en

Links

Abstract

The invention discloses a stereoscopic image objective quality evaluation method on the basis of the structural distortion, which comprises the following steps: firstly, respectively carrying out regional division on left and right viewpoint images of an undistorted stereoscopic image and a distorted stereoscopic image to obtain an eye sensitive region and a corresponding nonsensitive region and then respectively obtaining evaluation indexes of the sensitive region and the nonsensitive region from two aspects of the structural amplitude distortion and the structural direction distortion; secondly, acquiring quality evaluation values of the left and right viewpoint images; thirdly, sampling singular value difference and a mean deviation ratio of residual images of which singular values are deprived to evaluate the distortion condition of the depth perception of a stereoscopic image so as to obtain an evaluation value of stereoscopic perceived quality; and finally, combining the quality of the left and right viewpoint images with the stereoscopic perceived quality to obtain a final quality evaluation result of the stereoscopic image. The method disclosed by the invention avoids simulating each composition part of an eye vision system, but sufficiently utilizes structure information of the stereoscopic image, so the consistency of the objective evaluation result and the subjective perception is effectively improved.

Description

一种基于结构失真的立体图像客观质量评价方法 An objective quality assessment method of the stereoscopic image distortion based on the structure

技术领域[0001] 本发明涉及一种图像质量评价技术,尤其是涉及一种基于结构失真的立体图像客观质量评价方法。 Technical Field [0001] The present invention relates to an image quality evaluation technique, particularly to a structure of an objective quality assessment of the stereoscopic image distortion based approach.

背景技术 Background technique

[0002] 立体图像质量评价在立体图像/视频系统中占据着十分重要的地位,不仅能够评判立体图像/视频系统中处理算法的优劣,而且还能优化和设计该算法,以提高立体图像/视频处理系统的效率。 [0002] stereoscopic image quality evaluation occupies a very important position in the stereo image / video system, not only to judge the merits of a stereoscopic image / video processing algorithms of the system, but also the design and optimization algorithms to improve the three-dimensional image / efficiency of video processing system. 立体图像质量评价方法主要分为两类:主观质量评价和客观质量评价。 Stereoscopic image quality assessment methods are mainly divided into two categories: subjective and objective quality assessment quality assessment. 主观质量评价方法就是把多名观察者对待评价立体图像的质量进行加权平均的综合评价,其结果符合人眼视觉系统特性,但是其受到计算不便、速度慢、成本高等诸多因素的限制,导致嵌入系统难,因而在实际应用中无法得到广泛推广。 Subjective quality assessment method is to treat the viewer than the stereoscopic image quality evaluation, were evaluated by a weighted average, a result consistent with human visual system, but by calculating the inconvenience, slow speed, high cost limit many factors, resulting in embedding the system is difficult, and therefore can not be widely used in practical applications. 而客观质量评价方法具有操作简单、成本低、易于实现及实时优化算法等特点,成为立体图像质量评价研究的重点。 The objective quality evaluation method is simple, low cost, easy to implement and real-time optimization, etc., become the focus of the study stereoscopic image quality assessment.

[0003]目前,主流的立体图像客观质量评价模型包括左右视点图像质量评价和深度感知质量评价两部分。 [0003] Currently, the main objective of the stereoscopic image quality assessment model includes left and right viewpoint image quality evaluation and quality evaluation of depth perception in two parts. 但是,由于人类对人眼视觉系统的认识有限,难以准确地模拟人眼的各个组成部分,因此这些模型与主观感知之间的一致性不是很好。 However, due to limited human understanding of the human visual system, it is difficult to simulate the various components of the human eye accurately, so the consistency between these models and subjective perception is not very good.

发明内容 SUMMARY

[0004] 本发明所要解决的技术问题是提供一种基于结构失真的立体图像客观质量评价方法,其能够有效提高立体图像客观质量评价结果与主观感知之间的一致性。 [0004] The present invention solves the technical problem is to provide an objective quality assessment method of the stereoscopic image distortion based on the structure, which can effectively improve the consistency between the objective quality stereoscopic image perception and subjective evaluation results.

[0005] 本发明解决上述技术问题所采用的技术方案为:一种基于结构失真的立体图像客观质量评价方法,其特征在于包括以下步骤: [0005] aspect of the present invention to solve the above technical problem is: an objective quality assessment method of the stereoscopic image distortion based on the structure, characterized by comprising the steps of:

[0006]①令Sots为原始的无失真的立体图像,令Sdis为待评价的失真的立体图像,将原始的无失真的立体图像Sots的左视点灰度图像记为Lots,将原始的无失真的立体图像Sots的右视点灰度图像记为R„g,将待评价的失真的立体图像Sdis的左视点灰度图像记为Ldis,将待评价的失真的立体图像Sdis的右视点灰度图像记为Rdis ; [0006] ① order Sots original stereoscopic image without distortion, so that a stereoscopic image distortion Sdis to be evaluated, the original undistorted left-view stereoscopic image Sots denoted Lots grayscale image, the original undistorted right viewpoint gradation Sdis stereoscopic image is a stereoscopic image is a stereoscopic image Sots Sdis right view gradation image referred to as R "g, distortion is evaluated to be a left-viewpoint image gradation referred to as Ldis, distortion is evaluated to be the denoted Rdis;

[0007] ②对Lots和Ldis、Rorg和Rdis 4幅图像分别实施区域划分,分别得到LOTg和Ldis、Rorg和Rdis4幅图像各自对应的敏感区域矩阵映射图,将Lots和Ldis分别实施区域划分后得到的各自对应的敏感区域矩阵映射图的系数矩阵均记为\,对于\中坐标位置为(i,j)处的系数值,将其记为 [0007] ② obtained after Lots and Ldis, Rorg and Rdis 4 images are embodiments zoning, respectively LOTg and Ldis, Rorg and Rdis4 images respectively corresponding to the sensitive region of the matrix map, the Lots and Ldis each embodiment zoning each sensitive area of ​​the matrix coefficients of the matrix corresponding to the map are referred to as \ for \ the coordinate position (i, j) at the coefficient value, which is referred to as

Figure CN102708568AD00091

;区域,将R„g和!^分别实施区 ;! Area, the R "g ^ respectively embodiments and region

域划分后得到的各自对应的敏感区域矩阵映射图的系数矩阵均记为Ak,对于Ak中坐标位 Coefficient matrix corresponding to each matrix of sensitive areas map obtained by dividing each domain after referred to as Ak, Ak coordinates for the position

置为(i,j〉处的系数值,将其记为Vij, Coefficient value is set to (i, j> at the denoted as Vij,

Figure CN102708568AD00092

,其中,此处 Where here

(W-8), O ≤ j ≤ (H-8),W 表不Lorg^Ldi^Rorg 和Rdis 的宽,H 表不Larg、Ldis、ROTg 和Rdis的闻; (W-8), O ≤ j ≤ (H-8), W table does not Lorg ^ Ldi ^ Rorg and Rdis wide, H table does not Larg, Ldis, ROTg Rdis and smell;

[0008] ③将Lots和Ldis2幅图像分别分割成(W-7) X (H_7)个尺寸大小为8X8的重叠块,然后计算Lots和Ldis2幅图像中所有坐标位置相同的两个重叠块的结构幅值失真映射图,将该结构幅值失真映射图的系数矩阵记为4,对于&中坐标位置为(i,j)处的系数值,将其 [0008] ③ Lots and the images are divided into Ldis2 (W-7) X (H_7) overlapping blocks of a size of 8X8, and then calculates Lots Ldis2 all images in the same coordinate positions of two overlapping blocks of structure amplitude distortion map, the magnitude of the coefficient structure of the distortion map of FIG. 4 referred to as a matrix for the & coordinate position (i, j) at the coefficient values, which

2 X CT (/ /■) + ί1 2 X CT (/ / ■) + ί1

记为Bl (i,j),BL{iJ) = ~^,其中,Bl (i,j)亦表示LOTg 中左 Referred to as Bl (i, j), BL {iJ) = ~ ^, wherein, Bl (i, j) is also represented in the left LOTg

(a0rgjj,j)) +(σ似(U))- + C1 (A0rgjj, j)) + (σ-like (U)) - + C1

上角坐标位置为(i,j)的尺寸大小为8X8的重叠块与Ldis中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块的结构幅值失真值, Size Size upper corner coordinate position (i, j) is a 8X8 block overlaps the top left corner Ldis coordinate position (i, j) is the amplitude of the distortion values ​​of the overlap structure of 8X8 blocks, and

(/', J去i Σ ^org ^ +xJ+ y) - Uo,^L (h J'))2 , (/ ', J to i Σ ^ org ^ + xJ + y) - Uo, ^ L (h J')) 2,

V o4 7=0 V o4 7 = 0

Uorg,L (A ■/)=去Σ Σ Lorg 0 + y) , Uorg, L (A ■ /) = go Σ Σ Lorg 0 + y),

04 X=OJ=O 04 X = OJ = O

σdm.L (J, j) = -,Itt ΣΣ +·*■,·/+ V) - Udis,L (?,j)Y , σdm.L (J, j) = -, Itt ΣΣ + · * ■, · / + V) - Udis, L (, j) Y,?

^ \|64 ^ \ | 64

= TJ Σ Σ Ldis Q + xJ + y) > = TJ Σ Σ Ldis Q + xJ + y)>

64 ϊ=0 >I=o 64 ϊ = 0> I = o

^orgJisx(iJ) = 士ΣΣ((1-0 + xJ + y) — uOrgAiJ)) X (4,,(7' + XJ + j) ~ (?', i))) , ^ OrgJisx (iJ) = Disabled ΣΣ ((1-0 + xJ + y) - uOrgAiJ)) X (4 ,, (7 '+ XJ + j) ~ (?')),, I)

x=ov=o x = ov = o

Lorg(i+x, j+y)表示LOTg中坐标位置为(i+x,j+y)的像素点的像素值,Ldis(i+x, j+y)表示Ldis中坐标位置为(i+x, j+y)的像素点的像素值,C1表示常数,此处OS (W-8), Pixel value Lorg (i + x, j + y) represents LOTg the coordinate position (i + x, j + y) of the pixel point, Ldis (i + x, j + y) represents Ldis coordinate position (i pixel value + x, j + y) of the pixel point, a C1 represents a constant, where OS (W-8),

O ^ j ^ (H-8); O ^ j ^ (H-8);

[0009] 将ROTg和Rdis2幅图像分别分割成(W-7) X (H_7)个尺寸大小为8X8的重叠块,然后计算Rots和Rdis2幅图像中所有坐标位置相同的两个重叠块的结构幅值失真映射图,将该结构幅值失真映射图的系数矩阵记为Bk,对于Bk中坐标位置为(i,j)处的系数值,将其记 [0009] The ROTg Rdis2 and images are divided into (W-7) X (H_7) overlapping blocks of a size of 8X8, and then calculates Rots Rdis2 all images in the same coordinate positions of two overlapping blocks web structure a distortion value map, the magnitude of the distortion coefficient map structure referred to as a matrix Bk, Bk for the coordinate position (i, j) at the coefficient values, which will be referred to

为Bk(i,j),BR(i,j) = -——,其中,BR(i, j)亦表示Rots 中左上 Of Bk (i, j), BR (i, j) = ---, where, BR (i, j) is also represented in the upper left Rots

角坐标位置为(i,j)的尺寸大小为8X8的重叠块与Rdis中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块的结构幅值失真值, Size Size angular coordinate position (i, j) is a 8X8 block overlaps the top left corner Rdis coordinate position (i, j) is the amplitude of the distortion values ​​of the overlap structure of 8X8 blocks, and

Jorg,R(U)=ΣΣ(Rorg(i + x / + y)~Uorg.R(i,j))2 , Jorg, R (U) = ΣΣ (Rorg (i + x / + y) ~ Uorg.R (i, j)) 2,

y η y η

uorg'R (J, j) = ττΣΧ Krg (i + xJ + y) , uorg'R (J, j) = ττΣΧ Krg (i + xJ + y),

MX=Q y=0 MX = Q y = 0

Figure CN102708568AD00111

Rorg(i+x, j+y)表示ROTg中坐标位置为(i+x,j+y)的像素点的像素值,Rdis(i+x, j+y)表示Rdis中坐标位置为(i+x, j+y)的像素点的像素值,C1表示常数,此处O < (W-8), Pixel value Rorg (i + x, j + y) represents ROTg the coordinate position (i + x, j + y) of the pixel point, Rdis (i + x, j + y) represents Rdis coordinate position (i the pixel values ​​of + x, j + y) of the pixel point, a C1 represents a constant, where O <(W-8),

O 彡j く(H-8); O San ku j (H-8);

[0010] ④对Lots和Ldis2幅图像分别实施水平及垂直方向Sobel算子处理,分别得到Lots和Ldis2幅图像各自对应的水平方向梯度矩阵映射图和垂直方向梯度矩阵映射图,将Lots实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,„&,对于Ih,OT“中坐标位置为(i,j)处的系数值,将其 [0010] ④ of Lots and Ldis2 images each embodiment horizontal and vertical Sobel operator processing, respectively Lots and Ldis2 images corresponding to each horizontal gradient matrix map and the vertical gradient matrix map, the embodiment in the horizontal direction Lots horizontal gradient matrix coefficients corresponding to the map after a Sobel operator process referred to as matrix obtained Ih, "&, for Ih, OT" of the coordinate position (i, j) at the coefficient values, which

Figure CN102708568AD00112

将LOTg实施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv,OTg』,对于Iv, 中坐标位置为(i,j)处的系数值,将其记 The coefficients of the vertical direction of the gradient matrix map corresponding to the LOTg embodiment of the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, OTG "for Iv, the coordinate position (i, j) coefficient values ​​at which was credited

Figure CN102708568AD00113

将Ldis实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,dis丨对于Ih,disュ中坐标位置为(i,j)处的系数值,将其记为Ih, The Ldis embodiment horizontal Sobel operator in the horizontal direction coefficient corresponding gradient matrix map latter sub-treating the resulting matrix is ​​referred to as Ih, dis Shu for Ih, dis ュ the coordinate position (i, j) coefficient values ​​at which was referred to as Ih,

Figure CN102708568AD00114

施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv, H,对于Iv, din中坐标位置为(i,j)处的系数值,将其记为Iv, n (i,j), Vertical coefficient corresponding to a direction gradient matrix map after application the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, H, for Iv, DIN the coordinate position (i, j) coefficient values ​​at, which is referred to as Iv , n (i, j),

Figure CN102708568AD00115

Lorg(i+2,j + 1)、Lorg(i+2, j+2)、Lorg(i, j)、Lorg(i, j + 1)、Lorg(i, j+2)、Lorg(i + l, j+2)、Lorg (i+1, j)分别对应表示LOTg 中坐标位置为(i+2,j)、(i+2,j+l)、(i+2,j+2)、(i,j)、(i,j+l)、(i,j+2)、(i+1, j+2), (i+1, j)的像素点的像素值,Ldis (i+2,j)、Ldis(i+2,j+Ι)、Ldis (i+2, j+2)、Ldis (i, j)、Ldis (i, j+1)、Ldis (i, j+2)、Ldis (i+1, j+2)、Ldis (i+1, j)分别对应表示Ldis 中坐标位置为(i+2, j), (i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2)、(i+1, j)的像素点的像素值; Lorg (i + 2, j + 1), Lorg (i + 2, j + 2), Lorg (i, j), Lorg (i, j + 1), Lorg (i, j + 2), Lorg (i + l, j + 2), Lorg (i + 1, j) respectively represent LOTg the coordinate position (i + 2, j), (i + 2, j + l), (i + 2, j + 2 pixel value), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel point, Ldis ( i + 2, j), Ldis (i + 2, j + Ι), Ldis (i + 2, j + 2), Ldis (i, j), Ldis (i, j + 1), Ldis (i, j +2), Ldis (i + 1, j + 2), Ldis (i + 1, j) respectively represent Ldis the coordinate position (i + 2, j), (i + 2, j + 1), ( i + 2, j + 2), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel pixel values;

[0011 ] 对Rots和Rdis2幅图像分别实施水平及垂直方向Sobel算子处理,分别得到Rots和Rdis2幅图像各自对应的水平方向梯度矩阵映射图和垂直方向梯度矩阵映射图,将ROTg实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为、㈣丨对于^^^中坐标位置为(i,j)处的系数值,将其记 [0011] The Rots and Rdis2 images each embodiment horizontal and vertical Sobel operator processing, respectively horizontal gradient matrix map and the vertical gradient matrix map Rots and Rdis2 images respectively corresponding to the ROTg embodiment in the horizontal direction Sobel horizontal gradient matrix coefficients corresponding to the map obtained after treatment operator referred to as a matrix, iv ^^^ Shu for the coordinate position (i, j) at the coefficient values, which will be referred to

Figure CN102708568AD00121

将R„g实施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv,OTg,K,对于Iv,„g,K中坐标位置为(i,j)处的系数值,将其记 The R "g After embodiment of the vertical direction Sobel operator treating the resulting corresponding vertical gradient matrix map coefficient matrix denoted Iv, OTg, K, for Iv," g, K the coordinate position (i, j) at the coefficient values, which will be referred to

Figure CN102708568AD00122

将Rdis实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,dis,K,对于Ih,dis,K中坐标位置为(i,j)处的系数值,将其记为Ih, Horizontal coefficients corresponding gradient matrix map after Rdis embodiment horizontal Sobel operator treating the resulting matrix is ​​referred to as Ih, dis, K, for Ih, dis, K the coordinate position (i, j) coefficient values ​​at , which is referred to as Ih,

Figure CN102708568AD00123

施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv, dis,K,对于Iv, dis,E中坐标位置为(i,j)处的系数值,将其记为Iv, dis;E(i, j), Vertical coefficient corresponding to a direction gradient matrix map after application the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, dis, K, for Iv, dis, E in the coordinate position (i, j) coefficient values ​​at the which is referred to as Iv, dis; E (i, j),

Figure CN102708568AD00124

Rorg(i+2, j + 1)、Rorg(i+2, j+2)、Rorg(i, j)、Rorg(i, j + 1)、Rorg(i, j+2)、Rorg(i + 1, j+2)、Rorg(i+1, j)分别对应表示ROTg 中坐标位置为(i+2,j)、(i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2), (i+1, j)的像素点的像素值,Rdis(i+2,j)、Rdis(i+2,j+1)、Rdis (i+2, j+2)、Rdis (i, j)、Rdis (i, j+1)、Rdis (i, j+2)、Rdis (i+1, j+2)、Rdis (i+1, j)分别对应表示Rdis 中坐标位置为(i+2, j), (i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2)、(i+1, j)的像素点的像素值; Rorg (i + 2, j + 1), Rorg (i + 2, j + 2), Rorg (i, j), Rorg (i, j + 1), Rorg (i, j + 2), Rorg (i + 1, j + 2), Rorg (i + 1, j) respectively represent ROTg the coordinate position (i + 2, j), (i + 2, j + 1), (i + 2, j + 2 pixel value), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel point, Rdis ( i + 2, j), Rdis (i + 2, j + 1), Rdis (i + 2, j + 2), Rdis (i, j), Rdis (i, j + 1), Rdis (i, j +2), Rdis (i + 1, j + 2), Rdis (i + 1, j) respectively represent Rdis the coordinate position (i + 2, j), (i + 2, j + 1), ( i + 2, j + 2), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel pixel values;

[0012] ⑤计算Lots和Ldis2幅图像中所有坐标位置相同的两个重叠块的结构方向失真映射图,将该结构方向失真映射图的系数矩阵记为匕,对于El中坐标位置为(i,j)处的系数值,将其记为Eji,j), [0012] ⑤ calculation coefficient Lots Ldis2 images and the same structure in all directions of the two coordinate positions overlapping blocks of the distortion map, the distortion direction of the structure referred to as a matrix map dagger, El for the coordinate position (i, coefficient value j) at which was referred to as EJI, j),

Figure CN102708568AD00125

常数; constant;

[0013] 计算Rots和Rdis2幅图像中所有坐标位置相同的两个重叠块的结构方向失真映射图,将该结构方向失真映射图的系数矩阵记为Εκ,对于Ek中坐标位置为(i,j)处的系数值,将其记为EK(i,j), [0013] and the calculation Rots Rdis2 structural direction images in the same coordinate position of all the two overlapping blocks distortion map, the distortion map structural direction coefficient matrix denoted Εκ, Ek for the coordinate position (i, j ) at a coefficient value, which is referred to as EK (i, j),

Figure CN102708568AD00126

[0014] ⑥计算LOTg和Ldis的结构失真评价值,记为QS Ql=CO WQ111, JCO2XQnm, U [0014] ⑥ Ldis structure and calculation LOTg distortion evaluation value referred to as QS Ql = CO WQ111, JCO2XQnm, U

Figure CN102708568AD00131

其中, among them,

Figure CN102708568AD00132

O1表示Lots和Ldis中敏感区域的权重值,ω2表示Lots和Ldis中非敏感区域的权重值; Lots and O1 represents the weight Ldis sensitive region weight value, ω2 represent weight values ​​Lots and Central Africa Ldis sensitive areas;

[0015] 计算Rots和Rdis的结构失真评价值,记为 [0015] Calculation and Rdis Rots structural distortion evaluation value, denoted

Figure CN102708568AD00133
Figure CN102708568AD00134
Figure CN102708568AD00135

其中, among them,

Figure CN102708568AD00136

ί表示Rots和Rdis中敏感区域的权重值, ί represent weight values ​​and Rdis Rots sensitive region,

Figure CN102708568AD00137

表示Rmg和Rdis中非敏感区域的权重值; Rdis denotes right and Central Africa Rmg sensitive region weight value;

[0016] ⑦根据和Qk计算待评价的失真的立体图像Sdis相对于原始的无失真的立体图像Sots的空间频率相似度度量,记为 [0016] ⑦ calculated distortion to be evaluated according to the stereoscopic image and Qk Sdis spatial frequency relative to the original similarity measure Sots stereoscopic image without distortion, denoted

Figure CN102708568AD00138

,其中,β i表示Ql的权值; Wherein, β i denotes the weight of Ql;

[0017]⑧计算Lots和Rots的绝对差值图像,以矩阵形式表示为化7 [0017] ⑧ Calculation and Lots Rots absolute difference image, expressed in the form of a matrix 7

Figure CN102708568AD00139

,计 ,meter

算Ldis和Rdis的绝对差值图像,以矩阵形式表示为 And the absolute difference calculation Ldis image Rdis, expressed in a matrix form as

Figure CN102708568AD001310

其中 among them

Figure CN102708568AD001311

为取绝对值符号; For the absolute value symbol;

[0018] ⑨将 [0018] ⑨ will

Figure CN102708568AD001312

with

Figure CN102708568AD001313

幅图像分别分割成 Images are divided into

Figure CN102708568AD001314

个互不重叠的尺寸大小为8X8的 Nonoverlapping the size of the 8X8

±夹,然后对 ± clip, and then

Figure CN102708568AD001315

with

Figure CN102708568AD001316

幅图像中的所有块分别实施奇异值分解,得到对应的由其每个块的奇异值矩阵组成的奇异值映射图和£ほ对应的由其每个块的奇异值矩阵组成的奇异值映射图,将フ实施奇异值分解后得到的奇异值映射图的系数矩阵记为Gots,对于Gots中第η个块的奇异值矩阵中坐标位置为(p,q)处的奇异值,将其记为 All blocks were embodiment singular value decomposition of the images, obtained by singular value matrix corresponding to each block consisting of singular value map map consisting of singular value and corresponding £ Mizuho singular value matrix by each of the blocks the fu embodiment singular value map coefficient matrix referred to the singular value decomposition of Gots, for the singular value Gots first η th block matrix coordinate position (p, q) singular values ​​at which was referred to as

Figure CN102708568AD001317

,将实施奇异值分解后得到的奇异值映射图的系数矩阵记为Gdis,对于Gdis中第η个块的奇异值矩阵中坐标位置为(P,q)处的奇异值,将其记为G:0?,g),其中,Wui表示和 , Coefficient embodiment singular value map the singular value decomposition of the matrix referred to as Gdis, for the singular value matrix Gdis first η block the coordinate position (P, q) singular values ​​at which was referred to as G :? 0, g), which, Wui representation and

Figure CN102708568AD001318

的宽,Hui表示 Wide, Hui represents

和が的高, And ga high,

Figure CN102708568AD001319

[0019] ⑩计算 [0019] ⑩ calculated

Figure CN102708568AD001320

泣对应的奇异值映射图和 Weeping map corresponding to the singular values ​​and

Figure CN102708568AD001321

は对应的奇异值映射图的奇异值偏差评价 Singular value offset evaluation ha corresponding singular value map

值,记为K Value, denoted K

Figure CN102708568AD001322

其中, among them,

G:(AP)表示Gots中第η个块的奇异值矩阵中坐标位置为 G: (AP) represents the singular value Gots first block η coordinate position in the matrix

Figure CN102708568AD001323

处的奇异值, Singular Value at,

Figure CN102708568AD001324

表示Gdis中第η个块的奇异值矩阵中坐标位置为 Gdis represents the singular value of the matrix block of η coordinate position

Figure CN102708568AD001325

处的奇异值;[0020] ⑪对D:和Dg分别实施奇异值分解,分别得到和各自对应的2个正交矩阵和I个奇异值矩阵,将实施奇异值分解后得到的2个正交矩阵分别记为X org和VOTg,将Dgf实施奇异值分解后得到的奇异值矩阵记为0OTg,Zorg XOorgXVorg=D0J,将实施奇异值分解后得到的2个正交矩阵分别记为Xdis和Vdis,将/¾/实施奇异值分解后得到的奇异值矩阵记为0dis,Zdis XOdisXVdts 二D=; Singular values ​​at; [0020] ⑪ to D: Dg and embodiments are singular value decomposition, and each respectively corresponding to the two orthogonal matrices and singular value matrix I, the Example 2 after the orthogonal singular value decomposition of matrix are denoted as X org and VOTg, the singular value matrix Dgf embodiment denoted the singular value decomposition is obtained 0OTg, Zorg XOorgXVorg = D0J, the embodiment 2 after orthogonal matrix obtained by singular value decomposition are referred to as Xdis and Vdis, after the singular values ​​of the matrix / ¾ / singular value decomposition of the embodiments referred to as 0dis, Zdis XOdisXVdts two D =;

[0021] @分别计算D巧和2幅图像剥夺奇异值后的残留矩阵图,将剥夺奇异值 [0021] D @ clever and calculate residual matrix image of FIG. 2 after the deprivation of singular value, the singular value deprive

后的残留矩阵图记为XOTg,Xorg= X 0rgx A XVOTg,将剥夺奇异值后的残留矩阵图记为Xdis,Xdis=XdisX A XVdis,其中,A表示单位矩阵,A的大小与0OTg和Odis的大小一致; After the stamp is a residual matrix XOTg, Xorg = X 0rgx A XVOTg, would deprive the residual matrix singular value stamp Xdis, Xdis = XdisX A XVdis, wherein, A represents a unit matrix, and the size of A with 0OTg of Odis the same size;

Figure CN102708568AD00141

[0022] 计算Xorg和Xdis的均值偏差率,记 [0022] The calculation of average deviation Xorg and Xdis rate, denoted

Figure CN102708568AD00142

,其中,X Where, X

表示XOTg和Xdis中的像素点的横坐标,I表示Xorg和Xdis中的像素点的纵坐标; XOTg pixel abscissa and the point Xdis, I represents the ordinate of the pixel in Xorg and Xdis;

[0023] ⑭计算待评价的失真的立体图像Sdis相对于原始的无失真的立体图像Sots的立体 [0023] ⑭ Sdis distortion stereoscopic image to be evaluated is calculated relative to the original undistorted three-dimensional stereoscopic image Sots

感知评价度量,记为 Sensory evaluation metric, denoted

Figure CN102708568AD00143

其中,τ表示常数,用于调节K和〃在Qs中所 Where, [tau] represents a constant, and K for adjusting the 〃 as Qs

起的重要性; Since importance;

[0024] ⑮根据Qf和Qs,计算待评价的失真的立体图像Sdis的图像质量评价分值,记为Q, [0024] ⑮ according Qf and Qs is, the image quality of the stereoscopic image evaluation score calculated distortion Sdis to be evaluated, referred to as Q,

Figure CN102708568AD00144

,其中,P表示权重系数值。 Wherein, P represents a weight coefficient values ​​weight.

[0025] 所述的步骤②中Lots和Ldis各自对应的敏感区域矩阵映射图的系数矩阵\的获取过程为: [0025] The coefficients in the step ② and Ldis Lots matrix sensitive area map corresponding to each matrix \ acquisition process is:

[0026] ②_al、对Lots作水平及垂直方向Sobel算子处理,得到LOTg的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, n和Zv, n,然后计算Lots的梯度幅值图,记为Z11, [0026] ②_al, Lots of horizontal and vertical directions as a Sobel operator to give LOTg horizontal gradient image and a vertical direction gradient image, denoted as Zh, n and Zv, n, and then calculate the gradient magnitude Lots FIG. denoted Z11,

Figure CN102708568AD00145

,其中,Z11 (x, y)表示Z11 中坐标位置为(x, y)的 Wherein, Z11 (x, y) represents the coordinate position Z11 (x, y) of

像素点的梯度幅值,Zhj n (x, y)表示Zh, u中坐标位置为(x,y)的像素点的水平方向梯度值,Zu1(Xj)表示Zv, u中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡X彡W,I ^ y ^ Hi,此处W'表示Z11的宽,H'表示Z11的高; Gradient magnitude of pixels, Zhj n (x, y) represents Zh, u in the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zu1 (Xj) represents Zv, u in the coordinate position (x , the vertical gradient value y) of the pixel point, the I X San San W, I ^ y ^ Hi, where W 'denotes the width of Z11, H' represents the high-Z11;

[0027] ②_a2、对Ldis作水平及垂直方向Sobel算子处理,得到Ldis的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, 12和Zv, 12,然后计算Ldis的梯度幅值图,记为Z12, [0027] ②_a2, Ldis of horizontal and vertical directions for a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Ldis, denoted as Zh, 12 and Zv, 12, and then calculate the gradient magnitude Ldis FIG. denoted Z12,

Figure CN102708568AD00146

,其中,Z12 (X,y)表示Z12 中坐标位置为(x,y)的 Wherein, Z12 (X, y) represents the coordinate position Z12 (x, y) of

像素点的梯度幅值,zh,12(x,y)表示Zh,12中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,12(x,y)表示Zv,12中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W,I ^ y ^ Hi,此处W'表示Z12的宽,H'表示Z12的高; Gradient magnitude of pixels, zh, 12 (x, y) represents Zh, 12 the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zv, 12 (x, y) represents Zv, 12 in vertical gradient value of the pixel coordinate position (x, y) of, I x San San W, I ^ y ^ Hi, where W 'denotes the width of Z12, H' represents the high-Z12;

[0028] ②-a 3、 计算划分区域时所需的阈值T,Γ = αχ·Ε^χ(ΣΣζπ(χ,ァ)+ΣΣζη(λ·,>,)),其中,α 为常数,Z11Uj)表示Z11 中坐标 [0028] ②-a 3, required for the calculation of divided areas threshold T, Γ = αχ · Ε ^ χ (ΣΣζπ (χ, ASTON) + ΣΣζη (λ ·,>,)), where, α is a constant, Z11Uj ) represents the coordinate Z11

"へM X=O j=0 X=O y=Q "Understands M X = O j = 0 X = O y = Q

位置为(x,y)的像素点的梯度幅值,Z12 (X,y)表示Z12中坐标位置为(x,y)的像素点的梯度幅值; Gradient magnitude of the pixel position (x, y) of, Z12 (X, y) represents the coordinate position Z12 (x, y) of the pixel gradient magnitude;

[0029] ②_a4、将Z11中坐标位置为(i, j)的像素点的梯度幅值记为Z11 (i,j),将Z12中坐标位置为(i,j)的像素点的梯度幅值记为Z12 (i,j),判断Z11 (i,j)>T或Z12 (i,j)>T是否成立,如果成立,则确定LOTg和Ldis中坐标位置为(i,j)的像素点属于敏感区域,并令\(i,j)=l,否则,确定Lots和Ldis中坐标位置为(i,j)的像素点属于非敏感区域,并令AL(i,j)=0,其中,O 彡i 彡(W-8), O ^ j ^ (H-8); Gradient magnitude of pixels [0029] ②_a4, Z11 in the coordinate position (i, j) is referred to as Z11 (i, j), the coordinate position Z12 (i, j) of the pixel gradient magnitude referred to Z12 (i, j), is determined Z11 (i, j)> T or Z12 (i, j)> T is satisfied, if established, is determined LOTg and Ldis coordinate position (i, j) of pixels belonging to the sensitive areas, and so \ (i, j) = l, otherwise, determining Lots and Ldis coordinate position (i, j) of pixels belonging to the non-sensitive region, and let AL (i, j) = 0, wherein , O San San i (W-8), O ^ j ^ (H-8);

[0030] 所述的步骤②中Rots和Rdis各自对应的敏感区域矩阵映射图的系数矩阵Ak的获取过程为: [0030] The coefficients in the step ② Rots and Rdis are each sensitive area of ​​the matrix corresponds to a map acquisition process matrix Ak is:

[0031] ②_bl、对Rots作水平及垂直方向Sobel算子处理,得到Rots的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, 和Zv, ,然后计算Rtffg的梯度幅值图,记为Zn, [0031] ②_bl, Rots of horizontal and vertical directions as a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Rots, denoted as Zh, and Zv is,, and then calculate the gradient magnitude FIG Rtffg, denoted Zn,

Zrl (X, V) = ^[(Zhrl (X, y)f + (Zyrj (x, y)f,其中,Zrl (x,y)表示Zrl 中坐标位置为(x,y)的 Zrl (X, V) = ^ [(Zhrl (X, y) f + (Zyrj (x, y) f, wherein, Zrl (x, y) represents a coordinate ZRL position (x, y) of

像素点的梯度幅值,Zhj rl (x, y)表示Zh,中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,rt(x,y)表示Zv,ri中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡X彡W,I彡y彡H',此处W'表示Zrt的宽,H'表示Zrt的高; Gradient magnitude of pixels, Zhj rl (x, y) represents Zh, the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zv, rt (x, y) represents Zv, ri coordinate position vertical gradient value of the pixel (x, y) of, I X San San W, I y San San H ', where W' denotes the width Zrt, H 'represents the high Zrt;

[0032] ②_b2、对Rdis作水平及垂直方向Sobel算子处理,得到Rdis的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, r2和Zv,r2,然后计算Rdi s的梯度幅值图,记为Zr2, [0032] ②_b2, Rdis of horizontal and vertical directions for a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Rdis, denoted as Zh, r2 and Zv, r2, and then calculate the gradient magnitude FIG Rdi s , denoted Zr2,

Zr2(x,y) = ^J(Zh r2(x,y))2 + (Uxj))2,其中,Zr2 (x, y)表不Zr2 中坐标位置为(x, y)的 Zr2 (x, y) = ^ J (Zh r2 (x, y)) 2 + (Uxj)) 2, where, Zr2 (x, y) in the table does not Zr2 coordinate position (x, y) of

像素点的梯度幅值,Zhj r2(x, y)表示Zh, r2中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,r2(x,y)表示Zv,r2中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W,I彡y彡H',此处W'表示Zrf的宽,H'表示ん的高; Gradient magnitude of pixels, Zhj r2 (x, y) represents Zh, r2 the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zv, r2 (x, y) represents Zv, r2 coordinates vertical gradient value of the pixel position (x, y) of, I x san san W, I y san san H ', where W' denotes the width Zrf, H 'represents the high-san;

[0033] ②-b3、计算划分区域时所需的阈值T ', [0033] ②-b3, required for calculating the threshold value divided area T ',

I W' H' Wt H' I W 'H' Wt H '

T'= αχX(ΣΣ2H(-γ,ァ)+ ΣΣも2>')),其中,α 为常数,Zrl(x,y)表示Zrl 中坐标 T '= αχX (ΣΣ2H (-γ, ASTON) + ΣΣ 2 mo>')), where, [alpha] is a constant, Zrl (x, y) represents a coordinate ZRL

位置为(x,y)的像素点的梯度幅值,Zr2 (X,y)表示Zr2中坐标位置为(x,y)的像素点的梯度幅值; Gradient magnitude of the pixel position (x, y) of, Zr2 (X, y) represents the coordinate position Zr2 (x, y) of the pixel gradient magnitude;

[0034] ②_b4、将Z11中坐标位置为(i, j)的像素点的梯度幅值记为Zrt (i, j),将Zr2中坐标位置为(i,j)的像素点的梯度幅值记为Zu (i,j),判断Zrt (i,j)>T或Zrf (i,j)>T是否成立,如果成立,则确定ROTg和Rdis中坐标位置为(i,j)的像素点属于敏感区域,并令AK(i,j)=l,否则,确定Rots和Rdis中坐标位置为(i,j)的像素点属于非敏感区域,并令AK(i,j)=0,其中,O 彡i 彡(W-8), O ^ j ^ (H-8)。 Gradient magnitude of pixels [0034] ②_b4, Z11 in the coordinate position (i, j) is referred to as Zrt (i, j), the coordinate position Zr2 (i, j) of the pixel gradient magnitude referred to as Zu (i, j), is determined Zrt (i, j)> T or Zrf (i, j)> T is satisfied, if established, is determined ROTg and Rdis coordinate position (i, j) of pixels belonging to the sensitive areas, and so AK (i, j) = l, otherwise, determining Rots and Rdis coordinate position (i, j) of pixels belonging to the non-sensitive region, and so AK (i, j) = 0, wherein , O San San i (W-8), O ^ j ^ (H-8).

[0035] 所述的步骤⑦中β !的获取过程为: ! [0035] In the step ⑦ β acquisition process is:

[0036] ⑦-I、采用η幅无失真的立体图像建立其在不同失真类型不同失真程度下的失真立体图像集,该失真立体图像集包括多幅失真的立体图像,其中,η > I ;[0037] ⑦_2、利用主观质量评价方法获取失真立体图像集中的每幅失真的立体图像的平均主观评分差值,记为DMOS,DMOS=100-M0S,其中,MOS表示主观评分均值,DMOS e [O,100]; [0036] ⑦-I, [eta] using a stereoscopic image without distortion web stereoscopic image set to establish its distortion at different levels of distortion in different types of distortion, the distortion of a stereoscopic image set comprises a plurality of stereoscopic image distortion, wherein, η> I; [0037] ⑦_2, the subjective quality assessment method using the mean difference acquiring a stereoscopic image distortion subjective scoring each piece of stereoscopic image distortion set, referred to as DMOS, DMOS = 100-M0S, wherein, the MOS represents the mean subjective score, DMOS e [ O, 100];

[0038] ⑦_3、根据步骤①至步骤⑥的操作过程,计算失真立体图像集中的每幅失真的立体图像的左视点图像相对于对应的无失真的立体图像的左视点图像的敏感区域的评价值QnU和非敏感区域的评价值Qn ; [0038] ⑦_3, according to the process in step ① to step ⑥ calculates a distortion evaluation value of the sensitive area of ​​the left view image distortion-free stereoscopic image corresponding to the relative to the left-view image of the stereoscopic image of each piece of distorted perspective image set and the non-sensitive region QnU evaluation value Qn;

[0039] ⑦_4、采用数学拟合方法拟合失真立体图像集中失真的立体图像的平均主观评分差值DMOS和对应的Qu和Qn,从而获得β I值。 [0039] ⑦_4, mathematical fitting method fits a stereoscopic image distortion stereoscopic image concentration difference between the mean opinion score distortion and a corresponding DMOS Qu and Qn, thereby obtaining β I value.

[0040] 与现有技术相比,本发明的优点在于首先分别对无失真的立体图像和失真的立体图像的左视点图像和右视点图像进行区域划分,得到人眼敏感区域和相应的非敏感区域,然后从结构幅值失真和结构方向失真两方面分别得出敏感区域和非敏感区域的评价指标; [0040] Compared with the prior art, advantages of the present invention is that each of the first left-viewpoint image and a stereoscopic image without distortion distortion stereoscopic image and a right viewpoint image area division, and sensitive areas of the human eye to give the corresponding non-sensitive region, then the amplitude distortion and the distortion direction of the structure two structures were obtained, and evaluation of non-sensitive areas of the sensitive region;

其次采用线性加权分别得到左视点图像质量评价值和右视点图像质量评价值,进而得到左右视点图像质量评价值;再次根据奇异值可以较好的表征立体图像结构信息特性,采样奇异值差异和剥夺奇异值后的残余图像的均值偏差率来衡量立体图像的深度感知的畸变情况,从而获得立体感知质量的评价值;最后将左右视点图像质量和立体感知质量两者以非线性方式结合,得到立体图像的最终质量评价结果,由于本发明方法避免模拟人眼视觉系统的各个组成部分,而充分利用了立体图像的结构信息,因此有效地提高了客观评价结果与主观感知的一致性。 Second linear weighting respectively a left viewpoint image quality evaluation values ​​and the right-viewpoint image quality evaluation value, and further to obtain left and right viewpoint image quality evaluation value; The singular values ​​can again better characterize the structural information of the stereoscopic image characteristics, and the singular value of the sample deprivation the average deviation of the residual image after the singular values ​​of aberration to measure depth perception of a stereoscopic image, thereby obtaining the evaluation value of the perceptual quality perspective; Finally, the left and right viewpoint images and stereoscopic quality perceived quality of both the combined non-linear manner, to give a stereoscopic the final result of image quality evaluation, since the method of the present invention avoids the various components of the simulation of the human visual system, and the full use of structure information of a stereoscopic image, thus effectively improving the consistency of the results of objective evaluation of subjective perception.

附图说明 BRIEF DESCRIPTION

[0041] 图I为本发明方法的总体实现框图; [0041] I achieve overall block diagram of the method of the present invention;

[0042]图 2a 为Akko&Kayo (640 X 480)立体图像; [0042] Figure 2a is Akko & Kayo (640 X 480) a stereoscopic image;

[0043]图 2b 为Alt Moabit (1024X 768)立体图像; [0043] FIG 2b is Alt Moabit (1024X 768) a stereoscopic image;

[0044]图 2C 为Balloons(1024X768)立体图像; [0044] FIG 2C is a stereoscopic image Balloons (1024X768);

[0045]图 2d 为Door Flowers (1024X 768)立体图像; [0045] Figure 2d is a stereoscopic image Door Flowers (1024X 768);

[0046]图 2e 为Kendo (1024X 768)立体图像; [0046] FIG. 2e (1024X 768) as a stereoscopic image Kendo;

[0047]图 2f 为Leaving Laptop (1024X 768)立体图像; [0047] FIG. 2f is a Leaving Laptop stereoscopic image (1024X 768);

[0048]图 2g 为Lovebirdl (1024X768)立体图像; [0048] FIG Lovebirdl 2g is a stereoscopic image (1024X768);

[0049]图汍为 Newspaper(1024X768)立体图像; [0049] FIG Newspaper Wan is a stereoscopic image (1024X768);

[0050]图 2 i 为Xmas (640 X 480)立体图像; [0050] FIG. 2 i is a Xmas (640 X 480) a stereoscopic image;

[0051]图 2j 为Puppy (720X480)立体图像; [0051] FIG. 2j Puppy is a stereoscopic image (720X480);

[0052]图 2k 为Soccer2(720X480)立体图像; [0052] FIG. 2k stereoscopic image Soccer2 (720X480);

[0053]图 21 为Horse (480 X 270)立体图像; [0053] FIG 21 is a (480 X 270) Horse stereoscopic image;

[0054] 图3为本发明方法的左视点图像质量评价框图; Left viewpoint image quality evaluation block diagram [0054] FIG. 3 shows the inventive method;

[0055] 图4a为不同α和ω i下的左视点图像质量和主观感知质量之间的CC性能变化图; [0055] FIG 4a is a CC of FIG performance variations between a left viewpoint image quality subjectively perceived quality and at different α and ω i;

[0056] 图4b为不同α和ω i下的左视点图像质量和主观感知质量之间的SROCC性能变化图; [0056] Figure 4b is a variation of FIG SROCC performance between a left viewpoint image quality subjectively perceived quality and at different α and ω i;

[0057] 图4c为不同α和ω i下的左视点图像质量和主观感知质量之间的RMSE性能变化图; [0057] Figure 4c is a variation of FIG RMSE performance between image quality subjectively perceived quality and left view under different α and ω i;

[0058] 图5a为在GJ1=I的情况下,不同α下的左视点图像质量和主观感知质量之间的CC性能变化图; [0058] In the case of Figure 5a is a GJ1 = I, CC FIG performance variations between a left viewpoint image quality for different α and subjectively perceived quality;

[0059] 图5b为在GJ1=I的情况下,不同α下的左视点图像质量和主观感知质量之间的SROCC性能变化图; [0059] In the case of Figure 5b is the GJ1 = I, SROCC FIG performance variations between a left viewpoint image quality for different α and subjectively perceived quality;

[0060] 图5c为在GJ1=I的情况下,不同α下的左视点图像质量和主观感知质量之间的RMSE性能变化图; [0060] In the case of FIG. 5c is a GJ1 = I, RMSE performance variations between a left viewpoint image of FIG quality for different α and subjectively perceived quality;

[0061] 图6a为不同β ί下的左右视点图像质量和主观感知质量之间的CC性能变化图; [0061] Figure 6a is a CC of FIG performance variations between the left and right viewpoint images at different quality subjectively perceived quality and β ί;

[0062] 图6b为不同β ί下的左右视点图像质量和主观感知质量之间的SROCC性能变化图; [0062] Figure 6b is a variation of FIG SROCC performance between right and left viewpoint images at different quality subjectively perceived quality and β ί;

[0063] 图6c为不同β ί下的左右视点图像质量和主观感知质量之间的RMSE性能变化图; [0063] Figure 6c is a variation of FIG RMSE performance between the left and right viewpoint image quality and various β ί subjectively perceived quality;

[0064] 图7a为不同τ下的立体深度感知质量和主观感知质量之间的CC性能变化图; [0064] FIG. 7a FIG perceived performance variations between CC and subjective quality perceived quality stereo depth under different [tau];

[0065] 图7b为不同τ下的立体深度感知质量和主观感知质量之间的SROCC性能变化图; [0065] FIG. 7b sensing SROCC FIG performance variations between the mass and the mass was subjectively perceived stereo depth under different [tau];

[0066] 图7c为不同τ下的立体深度感知质量和主观感知质量之间的RMSE性能变化图; [0066] FIG 7c sensing FIG RMSE performance variations between subjectively perceived quality and the quality of stereo depth under different [tau];

[0067] 图8a为不同P下的立体图像质量和主观感知质量之间的CC性能变化图; [0067] Figure 8a is a perspective CC performance between different image quality subjectively perceived quality and P change map;

[0068] 图8b为不同P下的立体图像质量和主观感知质量之间的SROCC性能变化图; [0068] FIG 8b is a perspective SROCC performance between different image quality subjectively perceived quality and P change map;

[0069] 图8c为不同P下的立体图像质量和主观感知质量之间的RMSE性能变化图。 [0069] Figure 8c is a variation of FIG RMSE performance between a stereoscopic image quality for different P and subjective perceptual quality.

具体实施方式 Detailed ways

[0070] 以下结合附图实施例对本发明作进ー步详细描述。 [0070] Hereinafter, the present invention will be described into detail in conjunction with the accompanying drawings ー Step embodiment.

[0071] 本发明提出的一种基于结构失真的立体图像客观质量评价方法,其从结构失真的角度分别评价了左右视点图像质量和立体图像的立体感知质量,采用非线性加权的方式得到立体图像的最終质量评价值。 [0071] One proposed objective quality evaluation method of the present invention is a stereoscopic image based on distortion of the structure, which from a structural perspective distortion perceived quality were evaluated perspective left view image and a stereoscopic image quality, using a weighted nonlinear manner to obtain a stereoscopic image the final quality evaluation value. 图I给出了本发明方法的总体实现框图,其包括以下步骤: Figure I shows a block diagram of the overall method of the present invention is implemented, comprising the steps of:

[0072]①令Sots为原始的无失真的立体图像,令Sdis为待评价的失真的立体图像,将原始的无失真的立体图像Sots的左视点灰度图像记为Lots,将原始的无失真的立体图像Sots的右视点灰度图像记为R„g,将待评价的失真的立体图像Sdis的左视点灰度图像记为Ldis,将待评价的失真的立体图像Sdis的右视点灰度图像记为Rdis。 [0072] ① order Sots original stereoscopic image without distortion, so that a stereoscopic image distortion Sdis to be evaluated, the original undistorted left-view stereoscopic image Sots denoted Lots grayscale image, the original undistorted right viewpoint gradation Sdis stereoscopic image is a stereoscopic image is a stereoscopic image Sots Sdis right view gradation image referred to as R "g, distortion is evaluated to be a left-viewpoint image gradation referred to as Ldis, distortion is evaluated to be the denoted Rdis.

[0073] ②对Lorg和Ldis, Rorg和Rdis4幅图像分别实施区域划分,分别得到Lorg和Ldis, Rorg和Rdis4幅图像各自对应的敏感区域矩阵映射图,将Lots和Ldis分别实施区域划分后得到的各自对应的敏感区域矩阵映射图的系数矩阵均记为\,对于\中坐标位置为(i,j)处 [0073] ② of Lorg and Ldis, Rorg and Rdis4 images are embodiments zoning, respectively obtained after Lorg and Ldis, Rorg and Rdis4 images respectively corresponding to the sensitive region of the matrix map, the Lots and Ldis each embodiment divided areas each sensitive area of ​​the matrix coefficients of the matrix corresponding to the map are referred to as \ for \ the coordinate position (i, j) at

的系数值,将其记为 The coefficient values, which will be referred to as

Figure CN102708568AD00171

,将ROTg和Rdis分别实施区 , Respectively and Rdis regions embodiment ROTg

域划分后得到的各自对应的敏感区域矩阵映射图的系数矩阵均记为Ak,对于Ak中坐标位置为(i,j)处的系数值,将其记为 Coefficient matrix corresponding to each matrix of sensitive areas map obtained by dividing each domain after referred to as Ak, Ak for coefficient values ​​in the coordinate position at the (i, j), which is referred to as

Figure CN102708568AD00172

其中,此处 Among them, here

(W-8), O ≤ j ≤ (H-8) ,W 表不Lorg>Ldis>Rorg 和Rdis 的宽,H 表不Lorg、Ldis、Rorg 和Rdis的闻。 (W-8), O ≤ j ≤ (H-8), W table does not Lorg> Ldis> broad, H table does not Lorg, Ldis, Rorg and Rdis smell of Rorg and Rdis.

[0074] 在此具体实施例中,步骤②中Lots和Ldis各自对应的敏感区域矩阵映射图的系数矩阵Al的获取过程为: [0074] In this particular embodiment, the coefficients in step ② and Ldis Lots matrix sensitive area map retrieval process corresponding to each matrix of Al is:

[0075] ②_al、对Lots作水平及垂直方向Sobel算子处理,得到Lots的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, n和Zv, n,然后计算Lots的梯度幅值图,记为Z11, [0075] ②_al, Lots of horizontal and vertical directions as a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Lots, denoted as Zh, n and Zv, n, and then calculate the gradient magnitude Lots FIG. denoted Z11,

Figure CN102708568AD00181

,其中,Z11 (x, y)表示Z11 中坐标位置为(x,y)的 Wherein, Z11 (x, y) represents the coordinate position Z11 (x, y) of

像素点的梯度幅值,Zhj n (x, y)表示Zh, u中坐标位置为(x,y)的像素点的水平方向梯度值,Zu1(Xj)表示Zv, u中坐标位置为(x,y)的像素点的垂直方向梯度值,I≤X≤W, Gradient magnitude of pixels, Zhj n (x, y) represents Zh, u in the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zu1 (Xj) represents Zv, u in the coordinate position (x , in the vertical direction y) of the pixel value gradient, I≤X≤W,

I ^ y ^ Hi,此处W'表示Z11的宽,H'表示Z11的高。 I ^ y ^ Hi, where W 'denotes the width of Z11, H' represents the high Z11.

[0076] ②_a2、对Ldis作水平及垂直方向Sobel算子处理,得到Ldis的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, 12和Zv, 12,然后计算Ldis的梯度幅值图,记为Z12, [0076] ②_a2, Ldis of horizontal and vertical directions for a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Ldis, denoted as Zh, 12 and Zv, 12, and then calculate the gradient magnitude Ldis FIG. denoted Z12,

Figure CN102708568AD00182

,其中,Z12 (X,y)表示Z12 中坐标位置为(x,y)的 Wherein, Z12 (X, y) represents the coordinate position Z12 (x, y) of

像素点的梯度幅值,Zh,12(x,y)表示Zh,12中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,12(x,y)表示Zv,12中坐标位置为(x,y)的像素点的垂直方向梯度值,I≤x≤W, Gradient magnitude of pixels, Zh, 12 (x, y) represents Zh, 12 the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zv, 12 (x, y) represents Zv, 12 in vertical gradient value of the pixel coordinate position (x, y) of, I≤x≤W,

I ^ y ^ Hi,此处W'表示Z12的宽,H'表示Z12的高。 I ^ y ^ Hi, where W 'denotes the width of Z12, H' represents the high Z12.

[0077] ②-a 3、计算划分区域时所需的阈值T, [0077] ②-a 3, a desired divided area when calculating the threshold T,

Figure CN102708568AD00183

其中W 表示Z11 和Z12 的宽,H'表示Z11 Wherein W represents a wide Z11 and Z12, H 'represents Z11

和Z12的高,α为常数,Z11(XJ)表示Z11中坐标位置为(x,y)的像素点的梯度幅值,Z12 (x, y)表示Z12中坐标位置为(x,y)的像素点的梯度幅值。 And Z12 is high, α is a constant, Z11 (XJ) Z11 represents the coordinate position (x, y) of the pixel gradient magnitude, Z12 (x, y) represents the coordinate position Z12 (x, y) of gradient magnitude of pixels.

[0078] ②_a4、将Z11中坐标位置为(i,j)的像素点的梯度幅值记为Z11 (i,j),将Z12中坐标位置为(i,j)的像素点的梯度幅值记为Z12 (i,j),判断Z11 (i,j)>T或Z12 (i,j)>T是否成立,如果成立,则确定LOTg和Ldis中坐标位置为(i,j)的像素点属于敏感区域,并令\(i,j)=l,否则,确定Lots和Ldis中坐标位置为(i,j)的像素点属于非敏感区域,并令AL(i,j)=0,其中,O ≤ i ≤(W-8), O ^ j ^ (H-8)。 Gradient magnitude of pixels [0078] ②_a4, Z11 in the coordinate position (i, j) is referred to as Z11 (i, j), the coordinate position Z12 (i, j) of the pixel gradient magnitude referred to Z12 (i, j), is determined Z11 (i, j)> T or Z12 (i, j)> T is satisfied, if established, is determined LOTg and Ldis coordinate position (i, j) of pixels belonging to the sensitive areas, and so \ (i, j) = l, otherwise, determining Lots and Ldis coordinate position (i, j) of pixels belonging to the non-sensitive region, and let AL (i, j) = 0, wherein , O ≤ i ≤ (W-8), O ^ j ^ (H-8).

[0079] 在此具体实施例中,步骤②中Rots和Rdis各自对应的敏感区域矩阵映射图的系数矩阵Ak的获取过程为: [0079] In this particular embodiment, the coefficient matrix of sensitive areas map in step ② and Rdis Rots acquisition process corresponding to each matrix Ak is:

[0080] ②_bl、对Rots作水平及垂直方向Sobel算子处理,得到ROTg的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, 和Zv, ,然后计算Rtffg的梯度幅值图,记为Zn, [0080] ②_bl, Rots of horizontal and vertical directions as a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image ROTg, denoted as Zh, and Zv is,, and then calculate the gradient magnitude FIG Rtffg, denoted Zn,

Figure CN102708568AD00184

,其中,Zrl (x, y)表示Zrl 中坐标位置为(x,y)的 Wherein, Zrl (x, y) represents a coordinate ZRL position (x, y) of

像素点的梯度幅值,Zhj rl (x, y)表示Zh, 中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,rt(x,y)表示Zv,ri中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡X彡W, Gradient magnitude of pixels, Zhj rl (x, y) represents Zh, the coordinate position (x, y) in the horizontal direction of the gradient values ​​of pixel points, Zv, rt (x, y) represents Zv, ri coordinate position is (x, y) of pixels in the vertical direction of the gradient value, I X San San W,

I ^ y ^ Hi,此处W'表示Zrt的宽,H'表示Zrt的高。 I ^ y ^ Hi, where W 'denotes the width Zrt, H' represents the high Zrt.

[0081] ②-b2、对Rdis作水平及垂直方向Sobel算子处理,得到Rdis的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, r2和Zv,r2,然后计算Rdi s的梯度幅值图,记为Zr2, [0081] ②-b2, of Rdis for horizontal and vertical Sobel operator to give the horizontal gradient image and a vertical gradient image Rdis respectively referred to as a Zh, r2 and Zv, r2, and then calculate the gradient web Rdi s of FIG value, referred to as Zr2,

Figure CN102708568AD00185

,其中,Zr2(x,y)表示Zr2 中坐标位置为(x,y)的像素点的梯度幅值,Zhj r2(x, y)表示Zh, r2中坐标位置为(X,y)的像素点的水平方向梯度值,Zv,r2(x,y)表示Zv,r2中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W,I彡y彡H',此处W'表示Zrf的宽,H'表示ん的高。 Wherein, Zr2 (x, y) represents the coordinate position Zr2 (x, y) of the pixel gradient magnitude, Zhj r2 (x, y) represents Zh, r2 the coordinate position (X, y) of the pixel horizontal gradient value point, Zv, r2 (x, y) represents Zv, r2 the coordinate position (x, y) in the vertical direction of the gradient values ​​of pixel points, I San x San W, I San y San H ', where W 'denotes the width Zrf, H' represents the high-san.

[0082] ②-b3、计算划分区域时所需的阈值T ', [0082] ②-b3, required for calculating the threshold value divided area T ',

Figure CN102708568AD00191

其中,r 表示zrl 和zr2 的宽,H'表示zrl Wherein, r represents zrl zr2 and width, H 'represents zrl

和Zr2的高,α为常数,Zrl(x,y)表示Zrl中坐标位置为(x, y)的像素点的梯度幅值,Zr2 (x, y)表示Zr2中坐标位置为(x,y)的像素点的梯度幅值。 And Zr2 high, α is a constant, Zrl (x, y) represents a coordinate position in ZRL (x, y) of the pixel gradient magnitude, Zr2 (x, y) represents the coordinate position Zr2 (x, y ) is the gradient magnitude of pixels.

[0083] ②-b4、将Zrl中坐标位置为(i, j)的像素点的梯度幅值记为Zrl (i,j),将Zr2中坐标位置为(i,j)的像素点的梯度幅值记为Zu (i,j),判断Zrt (i,j)>T或Zrf (i,j)>T是否成立,如果成立,则确定ROTg和Rdis中坐标位置为(i,j)的像素点属于敏感区域,并令AK(i,j)=l,否则,确定Rots和Rdis中坐标位置为(i,j)的像素点属于非敏感区域,并令AK(i,j)=0,其中,O 彡i 彡(W-8), O ^ j ^ (H-8)。 Gradient magnitude of pixels [0083] ②-b4, in the ZRL coordinate position (i, j) is referred to as Zrl (i, j), the coordinate position Zr2 (i, j) the gradient of a pixel referred to as amplitude Zu (i, j), it is determined Zrt (i, j)> T or Zrf (i, j)> T is satisfied, if established, is determined and Rdis ROTg coordinate position (i, j) of pixels belonging to the sensitive areas, and so AK (i, j) = l, otherwise, determining Rots and Rdis coordinate position (i, j) of pixels belonging to the non-sensitive region, and so AK (i, j) = 0 wherein, i San San O (W-8), O ^ j ^ (H-8).

[0084] 在本实施例中,利用如图2a、图2b、图2c、图2d、图2e、图2f、图2g、图2h、图2i、图 [0084] In the present embodiment, using Figure 2a, to Figure 2b, 2c, the FIG. 2d, FIG. 2E, Fig. 2F, FIG. 2g, Fig 2h, 2i FIGS., FIG.

2j、图2k和图21所示的12幅无失真的立体图像建立其在不同失真类型不同失真程度下的失真立体图像集,失真类型包括JPEG压缩、JP2K压缩、高斯白噪声、高斯模糊和H264编码失真,且立体图像的左视点图像和右视点图像同时同程度失真,该失真立体图像集共包括312幅失真的立体图像,其中JPEG压缩的失真的立体图像共60幅,JPEG2000压缩的失真的立体图像共60幅,高斯白噪声失真的立体图像共60幅,高斯模糊失真的立体图像共60幅,H264编码失真的立体图像共72幅。 2j, 12 web stereoscopic image without distortion as shown in FIG. 21 and FIG 2k stereoscopic image set to establish its distortion at different levels of distortion in different types of distortion, the distortion types include JPEG compression, JP2K compression, white Gaussian noise, Gaussian blur, and H264 coding distortion, and the left-viewpoint image and a right view image of a stereoscopic image while distortion same level, the distortion of the stereoscopic image set including a total of stereoscopic image 312 distorted, which JPEG compression stereoscopic image distortion were 60, JPEG2000 compression distortion 60 were stereoscopic image, the stereoscopic image distortion Gaussian white noise were 60, the stereoscopic image distortion Gaussian blur were 60, the stereoscopic image co-H264 72 coding distortion. 对上述312幅立体图像进行如上所述的区域分割。 Region 312 of the above-described stereoscopic image segmentation as described above.

[0085] 在本实施例中,α值决定了敏感区域划分的准确程度,若取值过大,则敏感区域会被误认为非敏感区域,若取值过小,则非敏感区域会被误认为敏感区域,因此其值的确定过程与左视点图像质量或右视点图像质量对立体图像质量的贡献来決定。 [0085] In the present embodiment, the value of [alpha] determines the accuracy of the sensitive area division, if the value is too large, the sensitive region is non-sensitive region mistake, if the value is too small, the non-sensitive region can be mistaken that the sensitive region, the process of determining the value of the left-view or right-view picture quality contribution to image quality of the stereoscopic image quality is determined.

[0086] ③将Lots和Ldis2幅图像分别分割成(W-7) X (H_7)个尺寸大小为8X8的重叠块,然后计算Lots和Ldis2幅图像中所有坐标位置相同的两个重叠块的结构幅值失真映射图,将该结构幅值失真映射图的系数矩阵记为A,对于も中坐标位置为(i,j)处的系数值,将其 [0086] ③ Lots and the images are divided into Ldis2 (W-7) X (H_7) overlapping blocks of a size of 8X8, and then calculates Lots Ldis2 all images in the same coordinate positions of two overlapping blocks of structure amplitude distortion map, the distortion of the amplitude coefficient map structure referred to as a matrix a, mo for the coordinate position (i, j) at the coefficient values, which

记为 Denoted

Figure CN102708568AD00192

,其中,BL(i, j)亦表示Lots 中左 Which, BL (i, j) also said Lots left

上角坐标位置为(i,j)的尺寸大小为8X8的重叠块与Ldis中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块的结构幅值失真值, Size Size upper corner coordinate position (i, j) is a 8X8 block overlaps the top left corner Ldis coordinate position (i, j) is the amplitude of the distortion values ​​of the overlap structure of 8X8 blocks, and

Figure CN102708568AD00193
Figure CN102708568AD00201

Lorg (i+x, j+y)表示Lots中坐标位置为(i+x,j+y)的像素点的像素值,Ldis (i+x, j+y)表示Ldis中坐标位置为(i+x, j+y)的像素点的像素值,C1表示常数,C1是为了避免 Pixel value Lorg (i + x, j + y) represents Lots coordinate position (i + x, j + y) of the pixel point, Ldis (i + x, j + y) represents Ldis coordinate position (i pixel value + x, j + y) of the pixel point, C1 represents a constant, C1 to avoid

Figure CN102708568AD00202

的分母出现零的情况,在实际应用过程中可 The denominator appears zero, can be in the actual application process

取C1=O. 01,此处O ≤ i ≤(W-8), O ^ j ^ (H-8)。 Take C1 = O. 01, where O ≤ i ≤ (W-8), O ^ j ^ (H-8).

[0087] 在此,考虑图像的像素点之间的相关性,ー个尺寸大小为8X8的重叠块与它最相邻的左块或右块有7列重叠,同样,该8X8重叠块与它最相邻的上块或下块有7行重叠。 [0087] Here, considering the correlation between the pixels of the image, a ー overlapping blocks of size 8X8 with its nearest neighbor block having a left or right blocks 7 overlaps the same, overlapping blocks of the 8X8 it most adjacent upper or lower block has a block of 7 lines overlap.

[0088] 将Rots和Rdis2幅图像分别分割成(W-7) X (H_7)个尺寸大小为8X8的重叠块,然后计算Rots和Rdis2幅图像中所有坐标位置相同的两个重叠块的结构幅值失真映射图,将该结构幅值失真映射图的系数矩阵记为Bk,对于Bk中坐标位置为(i,j)处的系数值,将其记 [0088] The Rots Rdis2 and images are divided into (W-7) X (H_7) overlapping blocks of a size of 8X8, and then calculates Rots Rdis2 all images in the same coordinate positions of two overlapping blocks web structure a distortion value map, the magnitude of the distortion coefficient map structure referred to as a matrix Bk, Bk for the coordinate position (i, j) at the coefficient values, which will be referred to

为Bk(i, j), Of Bk (i, j),

Figure CN102708568AD00203

,其中,Bk(i, j)亦表不Rorg 中左上 Wherein, Bk (i, j) is also not Rorg upper left table

角坐标位置为(i,j)的尺寸大小为8X8的重叠块与Rdis中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块的结构幅值失真值, Size Size angular coordinate position (i, j) is a 8X8 block overlaps the top left corner Rdis coordinate position (i, j) is the amplitude of the distortion values ​​of the overlap structure of 8X8 blocks, and

Figure CN102708568AD00204

Rorg (i+x, j+y)表示ROTg中坐标位置为(i+χ,j+y)的像素点的像素值,Rdis (i+χ, j+y)表示Rdis中坐标位置为(i+x, j+y)的像素点的像素值,C1表示常数,C1是为了避免 Pixel value Rorg (i + x, j + y) represents ROTg the coordinate position (i + χ, j + y) of the pixel point, Rdis (i + χ, j + y) represents Rdis coordinate position (i pixel value + x, j + y) of the pixel point, C1 represents a constant, C1 to avoid

Figure CN102708568AD00211

分母出现零的情况,在实际应用过程中可 Denominator appears zero, may be in the actual application process

[0089] ④对Lots和Ldis2幅图像分别实施水平及垂直方向Sobel算子处理,分别得到Lots和Ldis2幅图像各自对应的水平方向梯度矩阵映射图和垂直方向梯度矩阵映射图,将Lots实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,„gj,对于Ih,OTgj中坐标位置为(i,j)处的系数值,将其 [0089] ④ of Lots and Ldis2 images each embodiment horizontal and vertical Sobel operator processing, respectively Lots and Ldis2 images corresponding to each horizontal gradient matrix map and the vertical gradient matrix map, the embodiment in the horizontal direction Lots horizontal gradient matrix coefficients corresponding to the map after a Sobel operator process referred to as matrix obtained Ih, "gj, for Ih, OTgj the coordinate position (i, j) at the coefficient values, which

Figure CN102708568AD00212

将LOTg实施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv,OTg丨对于Ιν,_Λ中坐标位置为(i,j)处的系数值,将其记为Iv, The coefficients of the vertical direction of the gradient matrix map corresponding to the LOTg embodiment of the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, OTg Shu for Ιν, _Λ the coordinate position (i, j) coefficient values ​​at which was credited as Iv,

Figure CN102708568AD00213

施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih, u对于Ih,disi中坐标位置为(i,j)处的系数值,将其记为Ih,dis, Applying horizontal Sobel operator corresponding to the horizontal direction coefficient gradient matrix map latter sub-treating the resulting matrix is ​​referred to as Ih, u for Ih, disi the coordinate position (i, j) coefficient values ​​at, which is referred to as Ih is, dis,

Figure CN102708568AD00214

垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv, dis, L,对于Iv, dis, L中坐标位置为(i,J')处的系数值,将其记为Iv, diS,L(i. j), Vertical coefficient corresponding to a direction gradient matrix map rear vertical Sobel operator treating the resulting matrix is ​​referred to as Iv, dis, L, a coefficient value for Iv, dis, L the coordinate position (i, J ') at the which is referred to as Iv, diS, L (i. j),

Figure CN102708568AD00215

Lorg(i+2, j + 1)、Lorg(i+2, j+2)、Lorg(i, j)、Lorg(i, j + 1)、Lorg(i, j+2)、Lorg(i + l, j+2)、Lorg(i+1, j)分别对应表示LOTg 中坐标位置为(i+2,j)、(i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2), (i+1, j)的像素点的像素值,Ldis(i+2,j)、Ldis(i+2,j+1)、Ldis (i+2, j+2)、Ldis (i, j)、Ldis (i, j+1)、Ldis (i, j+2)、Ldis (i+1, j+2)、Ldis (i+1, j)分别对应表示Ldis 中坐标位置为(i+2, j), (i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2)、(i+1, j)的像素点的像素值。 Lorg (i + 2, j + 1), Lorg (i + 2, j + 2), Lorg (i, j), Lorg (i, j + 1), Lorg (i, j + 2), Lorg (i + l, j + 2), Lorg (i + 1, j) respectively represent LOTg the coordinate position (i + 2, j), (i + 2, j + 1), (i + 2, j + 2 pixel value), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel point, Ldis ( i + 2, j), Ldis (i + 2, j + 1), Ldis (i + 2, j + 2), Ldis (i, j), Ldis (i, j + 1), Ldis (i, j +2), Ldis (i + 1, j + 2), Ldis (i + 1, j) respectively represent Ldis the coordinate position (i + 2, j), (i + 2, j + 1), ( i + 2, j + 2), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel pixel value.

[0090] 对Rots和Rdis2幅图像分别实施水平及垂直方向Sobel算子处理,分别得到Rots和Rdis2幅图像各自对应的水平方向梯度矩阵映射图和垂直方向梯度矩阵映射图,将Rots实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为、㈣丨对于^^^中坐标位置为(i,j)处的系数值,将其记 [0090] The Rots and Rdis2 images each embodiment horizontal and vertical Sobel operator processing, respectively horizontal gradient matrix map and the vertical gradient matrix map Rots and Rdis2 images respectively corresponding to the embodiment in the horizontal direction Rots Sobel horizontal gradient matrix coefficients corresponding to the map obtained after treatment operator referred to as a matrix, iv ^^^ Shu for the coordinate position (i, j) at the coefficient values, which will be referred to

Figure CN102708568AD00216

将R„g实施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv,OTg,K,对于Iv,„g,K中坐标位置为(i,j)处的系数值,将其记 The R "g After embodiment of the vertical direction Sobel operator treating the resulting corresponding vertical gradient matrix map coefficient matrix denoted Iv, OTg, K, for Iv," g, K the coordinate position (i, j) at the coefficient values, which will be referred to

Figure CN102708568AD00221

将Rdis实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,dis,K,对于Ih,dis,K中坐标位置为(i,j)处的系数值,将其记为Ih, Horizontal coefficients corresponding gradient matrix map after Rdis embodiment horizontal Sobel operator treating the resulting matrix is ​​referred to as Ih, dis, K, for Ih, dis, K the coordinate position (i, j) coefficient values ​​at , which is referred to as Ih,

Figure CN102708568AD00222

施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv, dis,R,对于Iv, dis,R中坐标位置为(i,j)处的系数值,将其记为Iv, dis,R(i,j), Vertical coefficient corresponding to a direction gradient matrix map after application the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, dis, R, for Iv, dis, R in the coordinate position (i, j) coefficient values ​​at the which is referred to as Iv, dis, R (i, j),

Figure CN102708568AD00223

Rorg(i+2, j + 1)、Rorg(i+2, j+2)、Rorg(i, j)、Rorg(i, j + 1)、Rorg(i, j+2)、Rorg(i + 1, j+2)、Rorg(i+1, j)分别对应表示ROTg 中坐标位置为(i+2,j)、(i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2), (i+1, j)的像素点的像素值,Rdis (i+2,j)、Rdis(i+2,j+Ι)、Rdis (i+2, j+2)、Rdis (i, j)、Rdis (i, j+1)、Rdis (i, j+2)、Rdis (i+1, j+2)、Rdis (i+1, j)分别对应表示Rdis 中坐标位置为(i+2, j), (i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2)、(i+1, j)的像素点的像素值。 Rorg (i + 2, j + 1), Rorg (i + 2, j + 2), Rorg (i, j), Rorg (i, j + 1), Rorg (i, j + 2), Rorg (i + 1, j + 2), Rorg (i + 1, j) respectively represent ROTg the coordinate position (i + 2, j), (i + 2, j + 1), (i + 2, j + 2 pixel value), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel point, Rdis ( i + 2, j), Rdis (i + 2, j + Ι), Rdis (i + 2, j + 2), Rdis (i, j), Rdis (i, j + 1), Rdis (i, j +2), Rdis (i + 1, j + 2), Rdis (i + 1, j) respectively represent Rdis the coordinate position (i + 2, j), (i + 2, j + 1), ( i + 2, j + 2), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel pixel value.

[0091] ⑤计算Lots和Ldis2幅图像中所有坐标位置相同的两个重叠块的结构方向失真映射图,将该结构方向失真映射图的系数矩阵记为も,对于も中坐标位置为(i,j)处的系数值,将其记为Eji,j), Coefficient [0091] ⑤ same calculated coordinate position and the structural direction Lots Ldis2 two overlapping images of all blocks distortion map, the distortion direction of the structure referred to as a matrix map mo, mo for the coordinate position (i, coefficient value j) at which was referred to as EJI, j),

Figure CN102708568AD00224

其中,C2表不常数,C2是为了避免 Wherein, C2 table is not constant, C2 in order to avoid

Figure CN102708568AD00225

零的情况,在实际应用过程中可取C2=O. 02。 Zero, preferably C2 = O. 02 in the actual application process.

[0092] 计算Rots和Rdis2幅图像中所有坐标位置相同的两个重叠块的结构方向失真映射图,将该结构方向失真映射图的系数矩阵记为Εκ,对于Ek中坐标位置为(i,j)处的系数值,将其记为EK(i,j), [0092] and the calculation Rots Rdis2 structural direction images in the same coordinate position of all the two overlapping blocks distortion map, the distortion map structural direction coefficient matrix denoted Εκ, Ek for the coordinate position (i, j ) at a coefficient value, which is referred to as EK (i, j),

Avoid

Figure CN102708568AD00226

的分母出现 Denominator appears

为零的情况,在实际应用过程中可取C2=O. 02。 The case of zero, preferably C2 = O. 02 in the actual application process.

[0093] ⑥计算Lots和Ldis的结构失真评价值,记为QS Ql=Co ^QniJco2XQmi,い [0093] ⑥ Ldis structure and calculation Lots distortion evaluation value referred to as QS Ql = Co ^ QniJco2XQmi, い

Figure CN102708568AD00231

O1表示Lots和Ldis中敏感区域的权重值,ω2表示Lots和Ldis中非敏感区域的权重值。 O1 denotes a weight value and Ldis Lots sensitive region, ω2 represent weight values ​​Lots and Central Africa Ldis sensitive areas.

[0094] 计算ROTg和Rdis的结构失真评价值,记为QK,Qe= ω / ,XQm, κ+ω / 2XQnm, Ε, [0094] Calculation and Rdis ROTg structural distortion evaluation value, denoted QK, Qe = ω /, XQm, κ + ω / 2XQnm, Ε,

Figure CN102708568AD00232

ω ' I表示Rots和Rdis中敏感区域的权重值,ω ' 2表示Rmg和Rdis中非敏感区域的权重值。 ω 'I represent weight values ​​and Rdis sensitive areas in Rots, ω' 2 represent weight values ​​and Rdis Central African Rmg sensitive areas.

[0095] 在本实施例中,图3给出了左视点图像质量评价的实现框图。 [0095] In the present embodiment, Figure 3 shows a block diagram of an implementation of the left view image quality evaluation. 利用如图2a至图21所示的12幅无失真的立体图像建立由312幅失真的立体图像构成的失真立体图像集,对这312幅失真的立体图像采用公知的主观质量评价方法进行主观质量评价,得到312幅失真的立体图像各自的平均主观评分差值(DMOS, Difference Mean Opinion Scores),即每幅失真的立体图像的主观质量评分值。 12 using a stereoscopic image without distortion as shown in FIG. 2a-21 established that the distorted stereo image by the stereo image set 312 composed of distortion, the subjective quality assessment method of this stereoscopic image distortion 312 using a known subjective quality evaluation, to obtain respective stereoscopic image difference mean Opinion score (DMOS, difference mean Opinion scores), i.e. subjective quality scores for each of the stereoscopic image distortion distortion web 312. DMOS为主观评分均值(MOS)和满分(100)的差值,即DM0S=100-M0S,因此,DMOS值越大表示失真的立体图像的质量越差,DMOS值越小表示失真的立体图像的质量越好,且DMOS的取值范围为[0,100]。 DMOS mean subjective score (MOS) and out (100) a difference, i.e. DM0S = 100-M0S, therefore, the greater the value of DMOS a perspective distortion of the image of poorer quality, DMOS smaller the value of a stereoscopic image distortion the better the quality, and the DMOS range is [0,100]. 另一方面,对上述312幅失真的立体图像按本发明方法步骤①至⑥计算得到每幅失真的立体图像相应的Qnu和Qnm, L ;然后采用QL=O1XQ1^(I-CO1) XQmu进行作四参数Logistic函数非线性拟合,得到α和O1值。 On the other hand, the above-described stereoscopic image 312 distorted by the method of the present invention steps ① to ⑥ calculated stereoscopic image distortion of each respective web Qnu and Qnm, L; then using QL = O1XQ1 ^ (I-CO1) XQmu performed for nonlinear four parameter Logistic fitting function to obtain the value of α and O1. 这里,利用评估图像质量评价方法的3个常用客观参量作为评价指标,即非线性回归条件下的Pearson 相关系数(Correlation Coefficient, CC)、Spearman 相关系数(SpearmanRank-Order Correlation Coefficient, SROCC)和均方误差系数(Rooted Mean SquaredError, RMSE), CC反映失真的立体图像评价函数这一客观模型的精度,SROCC反映客观模型与主观感知之间的单调性情况。 Here, with the assessment of the quality evaluation method of an image three usual objective parameters as an evaluation index, Pearson correlation coefficient (Correlation Coefficient, CC) in i.e. non-linear regression conditions, the Spearman correlation coefficient (SpearmanRank-Order Correlation Coefficient, SROCC) and the mean square error coefficients (Rooted Mean SquaredError, RMSE), stereoscopic image distortion evaluation CC reflecting the objective function of the accuracy of the model, SROCC reflect the monotonicity case between objective and subjective perception model. RMSE反映其预测的准确性。 RMSE reflect the accuracy of its predictions. CC和SROCC值越高说明立体图像客观评价方法与DMOS相关性越好,RMSE值越低说明立体图像客观评价方法与DMOS相关性越好。 And higher values ​​indicate CC SROCC objective assessment method DMOS stereoscopic image with better correlation, RMSE value lower objective assessment method described DMOS stereoscopic image with better correlation. 图4a给出了不同α和GJ1下的312幅立体图像的左视点图像质量和主观感知质量之间的CC性能变化,图4b给出了不同α和CO1下的312幅立体图像的左视点图像质量和主观感知质量之间的SROCC性能变化,图4c给出了不同α和Q1下的312幅立体图像的左视点图像质量和主观感知质量之间的RMSE性能变化,分析图4a、图4b和图4c,可知CC和SROCC会随着01值的变大而变大,而RMSE随着ω i值的变大而变小,说明左视点图像质量主要是由敏感区域的质量决定,而α值的改变对左视点图像质量与主观感知之间的性能影响不大。 Figure 4a shows the performance variations between the left CC view and the image quality subjectively perceived quality of a stereoscopic image 312 at different α and GJ1, Figure 4b shows a left-view image 312 under a stereoscopic image and various α CO1 SROCC performance variations between the subjective quality and perceived quality, Figure 4c shows RMSE performance variations between the left viewpoint image 312 quality stereoscopic image in different α and Q1 and subjective perceptual quality analysis Figures 4a, 4b, and 4c, the CC and SROCC be understood as the larger value 01 becomes greater, the RMSE as ω i becomes large and the value becomes smaller, indicating that the main left viewpoint image quality is determined by the mass-sensitive area, the value α the change has little effect performance between the left viewpoint image quality and subjective perception right. 图5a给出了在ω1=1、ω2=0情况下,不同α下的312幅立体图像的左视点图像质量和主观感知质量之间的CC性能变化;图5b给出了在ω1=1、ω2=0情况下,不同α下的312幅立体图像的左视点图像质量和主观感知质量之间的SROCC性能变化;图5c给出了在CO1=U ω2=0情况下,不同α下的312幅立体图像的左视点图像质量和主观感知质量之间的RMSE性能变化;分析图5a、图5b和图5c,可知CC、SR0CC和RMSE值都在百分位上波动,但都存在ー个峰值。 Figure 5a shows in ω1 = 1, ω2 = 0 case where, CC performance variations between the left viewpoint image 312 of the stereoscopic image quality at different α and subjectively perceived quality; Figure 5b shows in ω1 = 1, ω2 = 0 under the situation, SROCC performance variations between the left viewpoint and image quality subjectively perceived quality of a stereoscopic image 312 at different [alpha]; FIG 5c shows in CO1 = U ω2 = 0 case, the 312 different [alpha] RMSE performance variations between the left-viewpoint stereo image width and image quality subjectively perceived quality; analysis Figures 5a, 5b and 5C, seen CC, SR0CC and RMSE values ​​are fluctuations in the percentile, but there are peaks ー. 因此,在本实施例中,取ω1=1,α=2. I。 Accordingly, in the present embodiment, taken ω1 = 1, α = 2. I.

[0096] ⑦根据和Qk计算待评价的失真的立体图像Sdis相对于原始的无失真的立体图像Sots的空间频率相似度度量,记为Qf,9ρ=β1Χ9ι+(1-β1) XQk,其中,β i表示Ql的权值。 [0096] ⑦ Sdis stereoscopic image to be evaluated and calculating Qk distortion with respect to the spatial frequency of the original stereoscopic image without distortion Sots similarity metric, referred to as Qf, 9ρ = β1Χ9ι + (1-β1) XQk, wherein β i represents the weight of Ql.

[0097] 在此具体实施例中,步骤⑦中β ί的获取过程为: [0097] In this particular embodiment, the β ί acquired in step ⑦ the process:

[0098] ⑦-I、采用η幅无失真的立体图像建立其在不同失真类型不同失真程度下的失真立体图像集,该失真立体图像集包括多幅失真的立体图像,其中,η > I。 [0098] ⑦-I, [eta] using a stereoscopic image without distortion web stereoscopic image set to establish its distortion at different levels of distortion in different types of distortion, the distortion of a stereoscopic image set comprises a plurality of stereoscopic image distortion, wherein, η> I.

[0099] ⑦_2、利用主观质量评价方法获取失真立体图像集中的每幅失真的立体图像的平均主观评分差值,记为DM0S,DM0S=100-M0S,其中,MOS表示主观评分均值,DMOS e [O, 100]。 [0099] ⑦_2, the subjective quality assessment method using the mean difference acquiring a stereoscopic image distortion subjective scoring each piece of stereoscopic image distortion set, referred to as DM0S, DM0S = 100-M0S, wherein, the MOS represents the mean subjective score, DMOS e [ O, 100].

[0100] ⑦-3、根据步骤①至步骤⑥的操作过程,计算失真立体图像集中的每幅失真的立体图像的左视点图像相对于对应的无失真的立体图像的左视点图像的敏感区域的评价值QnU和非敏感区域的评价值。 [0100] ⑦-3, according to the process in step ① to step ⑥ calculated distortion set a left-view image of the stereoscopic image distortion of each piece of stereoscopic image with respect to the sensitive region of the left-view image distortion-free stereoscopic image corresponding and the evaluation value of the evaluation value QnU non-sensitive areas.

[0101] ⑦-4、采用数学拟合方法拟合失真立体图像集中失真的立体图像的平均主观评分差值DMOS和对应的Qu和从而获得β ί值。 [0101] ⑦-4, using mathematical fitting method of fitting a stereoscopic image distortion stereoscopic image concentration difference distortion DMOS and mean opinion score and thereby obtain the corresponding β ί Qu value.

[0102] 在本实施例中,β :决定了げ对立体图像质量的贡献,针对块效应,立体图像质量大概是左视点图像的质量和右视点图像的质量之和的一半,针对模糊失真,立体图像质量主要取决于质量较好的那个视点。 [0102] In the present embodiment, β: determines the contribution ge stereoscopic image quality, for deblocking, the stereoscopic image quality is probably the sum of the mass of the quality and the right-view image left viewpoint image half, the fuzzy distortion, stereoscopic image quality depends on the viewpoint of better quality. 由于该立体图像测试库的左视点图像和右视点图像同时受到同程度的失真,左视点图像的质量和右视点图像的质量变化不大,故P1变化对立体图像的主观性能影响不大。 Since the left-view image and a right-view image of the stereoscopic image test library at the same time by the same degree of distortion, mass change and the right viewpoint image quality left-view image is not, therefore P1 changes have little effect on the subjective performance of stereoscopic images. 首先对上述312幅失真的立体图像按本发明方法步骤①至⑥计算得到每幅失真的立体图像相应的げ和Qk,然后采用四參数拟合得到P1值。 First, the stereoscopic image 312 distorted by the method of the present invention steps ① to ⑥ calculated stereoscopic image distortion of each respective web ge and Qk, then a four-parameter fit to obtain the value of P1. 图6a给出了不同h下的左右视点图像的质量与主观感知质量之间的CC性能变化,图6a给出了不同 Figure 6a shows a CC performance variations between the left and right viewpoint images at different quality subjectively perceived quality and h, FIG. 6a shows the different

下的左右视点图像的质量与主观感知质量之间的SROCC性能变化,图6a给出了不同β i下的左右视点图像的质量与主观感知质量之间的RMSE性能变化,分析图6a、图6b和图6c,可知随着P1值的变化,CC、SROCC和RMSE值变化不大,在百分位上波动,但都存在峰值。 SROCC performance variations between the left and right viewpoint images in the quality subjectively perceived quality and, Figure 6a shows RMSE performance variations between the left and right viewpoint images at different β i subjectively perceived quality and quality analysis Figures 6a, 6b and FIG. 6C, with the changes seen values ​​P1, CC, SROCC and RMSE values ​​little change, the fluctuation in the percentile, but there is a peak. 这里,取^1=O. 5ο Here, take ^ 1 = O. 5ο

[0103] ⑧计算Lots和Rots的绝对差值图像用矩阵Ζ):表示,即化ブI,计算Ldis和Rdis的绝对差值图像用矩阵D=表示,即 [0103] ⑧ Lots calculated absolute difference between the image and Rots matrix Ζ): indicates that of STAB I, and calculates Ldis Rdis = absolute difference between the image represented by the matrix D, the

Figure CN102708568AD00241

,其中,“ II ”为取绝对值符号。 Wherein, "II" as the absolute value symbol.

[0104] ⑨将和P: 2幅图像分别分割成 [0104] ⑨ and the P: 2 images is divided into

Figure CN102708568AD00242

个互不重叠的尺寸大小为8X8的 Nonoverlapping the size of the 8X8

±夹,然后对和D= 2幅图像中的所有块分别实施奇异值分解,得到の;? ± clip, D = 2, and then all blocks in the images are embodiments singular value decomposition to obtain の;? 对应的由其每个块的奇异值矩阵组成的奇异值映射图和对应的由其每个块的奇异值矩阵组成的奇异值映射图,将DSf实施奇异值分解后得到的奇异值映射图的系数矩阵记为Gots,对于Gots中 Map corresponding to singular values ​​by the composition of matrix of singular values ​​of each block and the corresponding singular value map by each block consisting of singular value matrix, the singular value map DSf embodiment the singular value decomposition of the obtained Gots coefficient matrix denoted, for the Gots

第η个块的奇异值矩阵中坐标位置为(p,q)处的奇异值,将其记为(八め,将实施奇异值分解后得到的奇异值映射图的系数矩阵记为Gdis,对于Gdis中第η个块的奇异值矩阵中坐标位置为(P,q)处的奇异值,将其记为GI.S(A<?),其中,Wui表示/¾?和£)な的宽,Hui表示 Matrix of singular values ​​of η block the coordinate position of the singular values ​​(p, q) at which was referred to as (eight Circular, the embodiment coefficient singular value map the singular value decomposition of the matrix is ​​referred to as Gdis, for Gdis the matrix of singular values ​​of η th block coordinate position singular values ​​(P, q) at which was referred to as GI.S (a <?), wherein, represents Wui / ¾? and £) width of na , Hui expressed

n„ 和的高 n "and the high

Figure CN102708568AD00243

[0105] 在此,为了降低计算复杂度,一个8X8的块与它最相邻的左块或右块或上块或下块没有重复列或重复行,即块与块互不重叠。 [0105] Here, in order to reduce computational complexity, a 8X8 blocks most adjacent to its left or right block or blocks on the lower block or blocks without repeat repetition row or column, i.e., the blocks do not overlap.

[0106] ⑩计算对应的奇异值映射图和对应的奇异值映射图的奇异值偏差评价 [0106] ⑩ singular value corresponding to the deviation of the evaluation calculation map and the singular values ​​corresponding singular value map

值,记为K, Value, denoted K,

Figure CN102708568AD00251

其中, among them,

Figure CN102708568AD00252

表示Gots中第η个块的奇异值矩阵中坐标位置为(ρ,P)处的奇异值,G:(p,p)表示Gdis中第η个块的奇异值矩阵中坐标位置为(ρ,ρ)处的奇异值。 Represents the singular value matrix Gots first η block the coordinate position (ρ, P) singular values ​​at, G: (p, p) represents the singular value matrix Gdis first η block the coordinate position (ρ, ρ) at the singular values.

[0107]⑪对DSf和分别实施奇异值分解,分别得到£>1?和£@各自对应的2个正交矩 [0107] ⑪ each embodiment of DSf and singular value decomposition, respectively £> 1? £ @ and 2 respectively corresponding orthogonal moments

阵和I个奇异值矩阵,将实施奇异值分解后得到的2个正交矩阵分别记为χ org和VOTg,将DI实施奇异值分解后得到的奇异值矩阵记为0OTg,Zorg χOorg χ Vorg = Dlr将的实施奇异值分解后得到的2个正交矩阵分别记为χ dis和Vdis,将实施奇异值分解后得到的奇异值矩阵记为Odis, Zdis X ^dis X ^dis =Dlr。 Matrix and a singular value matrix I, the embodiment 2 after orthogonal matrix singular value decomposition are referred to as χ org and VOTg, the singular value matrix embodiment DI after referred to as the singular value decomposition 0OTg, Zorg χOorg χ Vorg = after the Dlr embodiment singular value decomposition to give two orthogonal matrix are denoted as χ dis and Vdis, the embodiment of the matrix of singular values ​​obtained by singular value decomposition referred to as Odis, Zdis X ^ dis X ^ dis = Dlr.

[0108] 分别计算DU和2幅图像剥夺奇异值后的残留矩阵图,将剥夺奇异值 [0108] FIG calculate the residual matrix DU image deprivation and the two singular value, the singular value deprive

后的残留矩阵图记为Xorg, Xorg=XorgX A XVots,将£造剥夺奇异值后的残留矩阵图记为Xdis,Xdis=XdisX Λ XVdis,其中,A表示单位矩阵,A的大小与0OTg和Odis的大小一致。 After the residual matrix stamp is Xorg, Xorg = XorgX A XVots, the residue matrix after the stamp made £ deprivation singular value Xdis, Xdis = XdisX Λ XVdis, wherein, A represents a unit matrix, and the size of 0OTg A Odis the same size.

[0109] 计算XOTg和Xdis的均值偏差率,记为φ: [0109] Calculation and Xdis XOTg mean deviation rate, referred to as φ:

Figure CN102708568AD00253

其中, among them,

X表示Xorg和Xdis中的像素点的横坐标,y表示Xorg和Xdis中的像素点的纵坐标。 X abscissa of the pixel in Xorg and Xdis, y represents the vertical coordinate of the pixel in Xorg and Xdis. ⑭计算待评价的失真的立体图像Sdis相对于原始的无失真的立体图像Sots的立体感知评价度量,记255 ⑭ distortion calculating a stereoscopic image to be evaluated with respect to Sdis original undistorted stereoscopic stereoscopic image Sots sensory evaluation metric, denoted 255

为Qs,a =Iog2其中,τ表示常数,用于调节K和在免中所起的重要性。 It is Qs, a = Iog2 where, [tau] represents a constant, and K for adjusting the importance of free plays.

[0110] 在本实施例中,首先求取上述312幅失真的立体图像和10幅无失真的立体图像的绝对差值图像,然后按照本发明方法的步骤⑧至⑬计算得到每幅失真的立体图像相应的K和炉。 [0110] In the present embodiment, the image of the first absolute difference is obtained stereoscopic image 312 distorted and undistorted 10 stereoscopic image, and follow the steps of the method of the present invention to ⑧ ⑬ perspective distortion is calculated for each web image corresponding to the K and the furnace. 在此,τ值大小决定了奇异值偏差与残留信息在深度感知评价中所起的重要性。 Here, τ determines the value of the size and importance of the singular value deviation residual information in depth perception evaluation plays. 图7a给出了在不同τ下312幅失真的立体图像的立体感知质量与主观感知之间的CC性能变化,图7b给出了在不同τ下312幅失真的立体图像的立体感知质量与主观感知之间的SROCC性能变化,图7c给出了在不同τ下312幅失真的立体图像的立体感知质量与主观感知之间的RMSE性能变化,图7a、图7b和图7c中τ在[-164]范围内变化,分析图7a、图7b和图7c,可知CC、SROCC和RMSE与τ的变化都存在一个极值,且位置大致相同,这里取τ = -8 ο FIG 7a shows a perspective CC performance variations between the perceived quality of a stereoscopic image at different τ 312 web distortion and subjective perception, FIG. 7b shows a perspective view of a stereoscopic image perceived quality τ 312 at different web distortion and subjective SROCC performance variations between the perceived, Figure 7c shows RMSE performance variations between the different web τ 312 a stereoscopic image distortion and subjective perception of perceived quality, Figures 7a, 7b and 7c, the [tau] [- 164] within the range of variation, analysis Figures 7a, 7b and 7C, can be seen, the RMSE, and CC SROCC τ variations are present one extreme, and substantially the same position, where τ = -8 ο take

[0111] ⑮根据Qf和Qs,计算待评价的失真的立体图像Sdis的图像质量评价分值,记为Q,Q=QfX (Qs) ρ,其中,P表示权重系数值。 Evaluation of image quality of the stereoscopic image Sdis [0111] ⑮ according Qf and Qs is, the distortion is calculated to be evaluated score, referred to as Q, Q = QfX (Qs) ρ, where, P represents the weight coefficient values.

[0112] 在本实施例中,对上述312幅失真的立体图像按本发明方法步骤①至⑭计算得到每幅失真的立体图像相应的Qf和Qs,然后采用Q=QfX (Qs) Ρ进行作四參数Logistic函数非线性拟合,得到P,P值决定了左右视点图像的质量和立体感知质量在立体图像质量中的贡献。 [0112] In the present embodiment, the above-described stereoscopic image 312 distorted by the method of the present invention, the step of ① to respective Qf and Qs ⑭ calculated for each web distorted stereo image, and using Q = QfX (Qs) Ρ be as nonlinear four parameter Logistic fitting function to obtain P, P value determines the contribution of the left and right viewpoint images and three-dimensional quality perceived quality of the stereoscopic image quality. Qf和Qs值都是随着立体图像失真程度加深而变小,故P值的取值范围为大于O。 Qs and Qf values ​​are deepened as a stereoscopic image distortion degree becomes small, so the range is greater than the value P O. 图8a给出了在不同P值下的312幅立体图像的质量与主观感知质量之间的CC性能变化,图8b给出了在不同P值下的312幅立体图像的质量与主观感知质量之间的SROCC性能变化,图8c给出了在不同P值下的312幅立体图像的质量与主观感知质量之间的RMSE性能变化,分析图8a、图Sb、图Sc,可知P值取得太大或太小都会影响立体图像质量客观评价模型与主观感知之间的一致性,随着P值变化情况下,CC、SROCC和RMSE值都存在极值点,且大致位置相同,这里取P =0. 3。 Figure 8a shows the performance variations between the CC 312 a stereoscopic image quality at various P values ​​and the subjective perception of quality, in FIG. 8b shows the mass 312 of the stereoscopic image P at different values ​​of the subjectively perceived quality SROCC performance changes between, Figure 8c shows RMSE performance variations between subjectively perceived quality and quality of a stereoscopic image 312 at different P values, the analysis of FIG. 8a, Sb, FIG Sc, seen P value achieve much or too small will affect the consistency between the stereoscopic image quality and objective evaluation of subjective perception model, with the P value changes, CC, SROCC and RMSE values ​​are extremum point, and the same approximate location, where P = 0 takes 3.

[0113] 分析本实施例得到的失真的立体图像的图像质量评价函数Q=QfX (Qs) °_3的最終评价结果与主观评分DMOS之间的相关性。 [0113] Image quality evaluation function Q stereoscopic image analysis of the present embodiment obtained in Example distortion = the correlation between the final evaluation results QfX (Qs) ° _3 with subjective scoring DMOS. 首先按本实施例得到的失真的立体图像的图像质量评价函数Q=QfX (Qs)0-3计算得到的最終立体图像质量评价结果的输出值Q,然后将输出值Q做四參数Logistic函数非线性拟合,最后得到立体客观评价模型与主观感知之间的性能指标值。 First image quality evaluation function Q stereoscopic image according to the present embodiment obtained in Example distortion = QfX (Qs) 0-3 calculates a final stereoscopic image quality evaluation results obtained output value Q, and the output value Q do four parameter Logistic Function nonlinear fitting, the last value of the performance index obtained between the three-dimensional model and objective evaluation of subjective perception. 这里,利用评估图像质量评价方法的4个常用客观參量作为评价指标,即CC、SR0CC、常值比率(Outlier Ratio,OR)、RMSE。 Here, the image quality evaluation methods evaluate the use of four common objective parameters as an evaluation index, i.e., CC, SR0CC, constant value of the ratio (Outlier Ratio, OR), RMSE. OR反映立体图像质量客观评级模型的离散程度,即所有失真立体图像中四參数拟合后的评价值与DMOS之间的差异大于某一阈值的失真立体图像数目所占比例。 OR reflect the degree of dispersion of the stereoscopic image quality objective evaluation model, i.e., the difference between the evaluation value and DMOS stereoscopic image distortion all four parameter fit is greater than a certain threshold percentage of the number of stereoscopic image distortion. 表I给出了评价性能CC、SROCC, OR和RMSE系数,由表I数据可见,按本实施例得到的失真的立体图像的图像质量评价函数Q=QfX (Qs)°_3计算得到的最终评价结果的输出值Q与主观评分DMOS之间的相关性是很高的,CC值和SROCC值都超过O. 92,RMSE值低于6. 5,表明客观评价结果与人眼主观感知的结果较为一致,说明了本发明方法的有效性。 Table I shows the performance evaluation CC, SROCC, OR and RMSE coefficient, seen from the data in Table I, the image quality evaluation function according to this embodiment of the stereoscopic image distortion obtained in Example Q = QfX (Qs) ° _3 final evaluation calculated the correlation between the output value Q of the result is highly subjective scoring DMOS, the CC value and the value of more than SROCC O. 92, RMSE value is less than 6.5, the results indicate that the results of objective evaluation of subjective perception of the human eye is more consistent illustrate the effectiveness of the method of the present invention. 表I本实施得到的失真的立体图像的图像质量评价分值与主观评分之间的相关性 Correlation between the image quality distortion evaluation value table I of the present embodiment of the stereoscopic image obtained with subjective scoring

[0114] [0114]

Figure CN102708568AD00261

Claims (3)

1. 一种基于结构失真的立体图像客观质量评价方法,其特征在于包括以下步骤: ①令Sots为原始的无失真的立体图像,令Sdis为待评价的失真的立体图像,将原始的无失真的立体图像Sots的左视点灰度图像记为Lots,将原始的无失真的立体图像Sots的右视点灰度图像记为R„g,将待评价的失真的立体图像Sdis的左视点灰度图像记为Ldis,将待评价的失真的立体图像Sdis的右视点灰度图像记为Rdis ; ②对Lots和Ldis、ROTg和Rdis4幅图像分别实施区域划分,分别得到Lots和Ldis、ROTg和Rdis4幅图像各自对应的敏感区域矩阵映射图,将Lots和Ldis分别实施区域划分后得到的各自对应的敏感区域矩阵映射图的系数矩阵均记为\,对于\中坐标位置为(i,j)处的系数值,将其记为 An objective quality assessment method of the stereoscopic image distortion based on the structure, characterized by comprising the following: ① From the original order Sots stereoscopic image without distortion, so that a stereoscopic image distortion Sdis to be evaluated, the original undistorted left-view stereoscopic image gradation Sdis perspective left view image Sots gradation image referred to as Lots, undistorted original stereoscopic right view image Sots gradation image referred to as R "g, distortion is to be evaluated Ldis referred to, to be evaluated Sdis stereoscopic image distortion is referred to as a right viewpoint image gradation Rdis; ② of Lots and Ldis, ROTg embodiment and Rdis4 images are divided into regions, respectively Lots and Ldis, ROTg images and Rdis4 each corresponding coefficient matrix sensitive area of ​​the matrix map after each corresponding sensitive area of ​​the matrix map, the Lots and Ldis each embodiment zoning obtained are referred to as \ for \ the coordinate position (i, j) coefficients at value, which is referred to as
Figure CN102708568AC00021
,将Rots和Rdis分别实施区域划分后得到的各自对应的敏感区域矩阵映射图的系数矩阵均记为Ak,对于Ak中坐标位置为(i,j)处的系数值,将其记为 Coefficient matrix corresponding to each sensitive area of ​​the matrix map, the region Rots and Rdis are obtained by dividing embodiments are denoted by Ak, Ak for coefficient values ​​in the coordinate position at the (i, j), which is referred to as
Figure CN102708568AC00022
、,其中,此处O≤i≤(W-8), 和Rdis的宽,H表示Lorg、Ldis、Rorg 和Rdis 的高; ③将Lots和Ldis2幅图像分别分割成(W-7) X (H-7)个尺寸大小为8X8的重叠块,然后计算Lots和Ldis2幅图像中所有坐标位置相同的两个重叠块的结构幅值失真映射图,将该结构幅值失真映射图的系数矩阵记为4,对于&中坐标位置为(i,j)处的系数值,将其记为 ,, wherein, where O≤i≤ (W-8), and the width Rdis, H represents Lorg Ldis, Rorg and Rdis high,; ③ Lots and the images are divided into Ldis2 (W-7) X ( H-7) th size of 8X8 overlapping blocks, and then calculates and Ldis2 images Lots structural amplitude distortion map all the same coordinate position of the two overlapping blocks, the magnitude of the coefficient matrix referred to the structure of the distortion map of FIG. 4, the coordinate position for the & (i, j) at the coefficient value, which is referred to as
Figure CN102708568AC00023
亦表示、中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块与Ldis中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块的结构幅值失真值, Also said, in the top left coordinate position (i, j) as overlapping blocks of size 8X8 Ldis the top left corner coordinate position (i, j) is the size of the structure of the amplitude distortion value of overlapping blocks of 8X8,
Figure CN102708568AC00024
Lorg(i+x, j+y)表示LOTg中坐标位置为(i+x,j+y)的像素点的像素值,Ldis(i+x, j+y)表示Ldis中坐标位置为(i+x,j+y)的像素点的像素值,C1表示常数,此处(W-8),O≤ j≤(H-8); 将Rots和Rdis2幅图像分别分割成(W-7) X (H-7)个尺寸大小为8X8的重叠块,然后计算Rots和Rdis2幅图像中所有坐标位置相同的两个重叠块的结构幅值失真映射图,将该结构幅值失真映射图的系数矩阵记为Bk,对于Bk中坐标位置为(i,j)处的系数值,将其记为 Pixel value Lorg (i + x, j + y) represents LOTg the coordinate position (i + x, j + y) of the pixel point, Ldis (i + x, j + y) represents Ldis coordinate position (i + x, the pixel values ​​of j + y) of the pixel point, a C1 represents a constant, where (W-8), O≤ j≤ (H-8); Rots, and the images are divided into Rdis2 (W-7) X (H-7) th size of 8X8 overlapping blocks, and then calculates and Rdis2 images Rots structural amplitude distortion map all the same coordinate position of the two overlapping blocks, the structure of the amplitude coefficient of the distortion map of FIG. matrix denoted Bk, Bk for the position coordinates (i, j) at the coefficient value, which is referred to as
Figure CN102708568AC00031
其中,BR(i,j)亦表示Rorg中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块与Rdis中左上角坐标位置为(i,j)的尺寸大小为8X8的重叠块的结构幅值失真值, Wherein, BR (i, j) in the upper left corner Rorg also said coordinate position (i, j) as overlapping blocks of size 8X8 Rdis the top left corner coordinate position (i, j) of the size of the overlapping 8X8 amplitude distortion value of a block structure,
Figure CN102708568AC00032
Rorg(i+x, j+y)表示ROTg中坐标位置为(i+x,j+y)的像素点的像素值,Rdis(i+x, j+y)表示Rdis中坐标位置为(i+x,j+y)的像素点的像素值,C1表示常数,此处(W-8),0≤ j ≤(H-8); ④对Lots和Ldis2幅图像分别实施水平及垂直方向Sobel算子处理,分别得到Lots和Ldis2幅图像各自对应的水平方向梯度矩阵映射图和垂直方向梯度矩阵映射图,将Lots实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,„g丨对于Ih,中坐标位置为(i,j)处的系数值,将其 Pixel value Rorg (i + x, j + y) represents ROTg the coordinate position (i + x, j + y) of the pixel point, Rdis (i + x, j + y) represents Rdis coordinate position (i pixel value + x, j + y) of the pixel point, a C1 represents a constant, where (W-8), 0≤ j ≤ (H-8); ④ Lots of images and Ldis2 embodiment respectively horizontal and vertical Sobel operator handling, respectively Lots and Ldis2 images corresponding to each horizontal gradient matrix map and the vertical gradient matrix map, the embodiment horizontally Lots coefficient corresponding to the horizontal direction of the gradient matrix map after the sub-processing obtained Sobel operator matrix referred to as Ih, "g Shu coefficient values ​​for Ih is, the coordinate position (i, j) at which was
Figure CN102708568AC00033
将Lots实施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv,OTg,y对于Iv,OTgj中坐标位置为(i,j)处的系数值,将其记 The Lots embodiment of the vertical direction Sobel operator perpendicular coefficient corresponding to the direction of the gradient matrix of the map after the sub-treating the resulting matrix is ​​referred to as Iv, OTg, y coefficient value of (i, j) at for Iv, OTgj coordinate position, which remember
Figure CN102708568AC00034
将Ldis实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Im,对于Ih,disi中坐标位置为(i,j)处的系数值,将其记为Ih, The Ldis embodiment horizontal Sobel operator in the horizontal direction coefficient corresponding gradient matrix map latter sub-treating the resulting matrix is ​​referred to as Im, for Ih, DISI the coordinate position (i, j) coefficient values ​​at, which is referred to as Ih ,
Figure CN102708568AC00041
施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv, n,对于Iv, dis,L中坐标位置为(i,j)处的系数值,将其记为Iv, n (i,j), Vertical coefficient corresponding to a direction gradient matrix map after application the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, n, for Iv, dis, L the coordinate position (i, j) coefficient values ​​at which was credited is Iv, n (i, j),
Figure CN102708568AC00042
,其中, ,among them,
Figure CN102708568AC00043
Lorg(i+2, j + 1)、Lorg(i+2, j+2)、Lorg(i, j)、Lorg(i, j + 1)、Lorg(i, j+2)、Lorg(i + l, j+2)、Lorg(i+1, j)分别对应表示LOTg 中坐标位置为(i+2,j)、(i+2,j+l)、(i+2,j+2)、(i,j)、(i,j+l)、(i,j+2)、(i+1, j+2), (i+1, j)的像素点的像素值,Ldis(i+2,j)、Ldis (i+2,j+1)、Ldis (i+2, j+2)、Ldis (i, j)、Ldis (i, j+1)、Ldis (i, j+2)、Ldis (i+1, j+2)、Ldis (i+1, j)分别对应表示Ldis 中坐标位置为(i+2, j), (i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2)、(i+1, j)的像素点的像素值; 对Rwg和Rdis2幅图像分别实施水平及垂直方向Sobel算子处理,分别得到Rorg和Rdis2幅图像各自对应的水平方向梯度矩阵映射图和垂直方向梯度矩阵映射图,将Rots实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为、㈣丨对于^^^中坐标位置为(i,j)处的系数值,将其记 Lorg (i + 2, j + 1), Lorg (i + 2, j + 2), Lorg (i, j), Lorg (i, j + 1), Lorg (i, j + 2), Lorg (i + l, j + 2), Lorg (i + 1, j) respectively represent LOTg the coordinate position (i + 2, j), (i + 2, j + l), (i + 2, j + 2 pixel value), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel point, Ldis ( i + 2, j), Ldis (i + 2, j + 1), Ldis (i + 2, j + 2), Ldis (i, j), Ldis (i, j + 1), Ldis (i, j +2), Ldis (i + 1, j + 2), Ldis (i + 1, j) respectively represent Ldis the coordinate position (i + 2, j), (i + 2, j + 1), ( i + 2, j + 2), (i, j), (i, j + l), (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel the pixel values ​​of the points; for Rwg and Rdis2 images each embodiment horizontal and vertical Sobel operator processing, respectively Rorg and Rdis2 images corresponding to each horizontal gradient matrix map and the vertical gradient matrix map, the level of implementation of Rots horizontal gradient matrix coefficients corresponding to the map after a Sobel operator orientation matrix obtained referred to as a process, iv ^^^ Shu for the coordinate position (i, j) at the coefficient values, which will be referred to
Figure CN102708568AC00044
将R„g实施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv,OTg,K,对于Iv,„g,K中坐标位置为(i,j)处的系数值,将其记 The R "g After embodiment of the vertical direction Sobel operator treating the resulting corresponding vertical gradient matrix map coefficient matrix denoted Iv, OTg, K, for Iv," g, K the coordinate position (i, j) at the coefficient values, which will be referred to
Figure CN102708568AC00045
将Rdis实施水平方向Sobel算子处理后得到的对应的水平方向梯度矩阵映射图的系数矩阵记为Ih,dis,K,对于Ih,dis,K中坐标位置为(i,j)处的系数值,将其记为Ih, Horizontal coefficients corresponding gradient matrix map after Rdis embodiment horizontal Sobel operator treating the resulting matrix is ​​referred to as Ih, dis, K, for Ih, dis, K the coordinate position (i, j) coefficient values ​​at , which is referred to as Ih,
Figure CN102708568AC00046
施垂直方向Sobel算子处理后得到的对应的垂直方向梯度矩阵映射图的系数矩阵记为Iv, dis,K,对于Iv, dis,E中坐标位置为(i,j)处的系数值,将其记为Iv, dis;E(i, j),J) = ^dis i + 2) + IRdjs (i + 1,7+2) + Rdis (i + 2,] + 2) , ,.,其中,Ui + 2, j )、 -Rdis (U) — 2Rdls (i +1,j) — Rdts (i + 2j)Rorg(i+2, j + 1)、Rorg(i+2, j+2)、Rorg(i, j)、Rorg(i, j + 1)、Rorg(i, j+2)、Rorg(i + 1, j+2)、Rorg(i+1, j)分别对应表示ROTg 中坐标位置为(i+2,j)、(i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2), (i+1, j)的像素点的像素值,Rdis(i+2,j)、Rdis (i+2,j+1)、Rdis (i+2, j+2)、Rdis (i, j)、Rdis (i, j+1)、Rdis (i, j+2)、Rdis (i+1, j+2)、Rdis (i+1, j)分别对应表示Rdis 中坐标位置为(i+2, j), (i+2, j+1), (i+2, j+2), (i,j)、(i,j+l)、(i,j+2)、(i+1, j+2)、(i+1, j)的像素点的像素值; ⑤计算Lots和Ldis2幅图像中所有坐标位置相同的两个重叠块的结构方 Vertical coefficient corresponding to a direction gradient matrix map after application the vertical direction Sobel operator treating the resulting matrix is ​​referred to as Iv, dis, K, for Iv, dis, E in the coordinate position (i, j) coefficient values ​​at the which is referred to as Iv, dis; E (i, j), J) = ^ dis i + 2) + IRdjs (i + 1,7 + 2) + Rdis (i + 2,] + 2),,, therein. , Ui + 2, j), -Rdis (U) - 2Rdls (i + 1, j) - Rdts (i + 2j) Rorg (i + 2, j + 1), Rorg (i + 2, j + 2) , Rorg (i, j), Rorg (i, j + 1), Rorg (i, j + 2), Rorg (+ 1, j + 2), Rorg (i + 1, j) respectively represent ROTg in coordinate position (i + 2, j), (i + 2, j + 1), (i + 2, j + 2), (i, j), (i, j + l), (i, j + pixel value 2), (i + 1, j + 2), (i + 1, j) of the pixel point, Rdis (i + 2, j), Rdis (i + 2, j + 1), Rdis (i +2, j + 2), Rdis (i, j), Rdis (i, j + 1), Rdis (i, j + 2), Rdis (i + 1, j + 2), Rdis (i + 1, j) respectively represent Rdis the coordinate position (i + 2, j), (i + 2, j + 1), (i + 2, j + 2), (i, j), (i, j + l ), the pixel value (i, j + 2), (i + 1, j + 2), (i + 1, j) of the pixel points; ⑤ and calculating Lots Ldis2 all images in the same coordinate positions of two superimposed structure of a block 失真映射图,将该结构方向失真映射图的系数矩阵记为匕,对于El中坐标位置为(i,j)处的系数值,将其记为Eji,j), A distortion map, the map structural direction distortion coefficient matrix referred to as dagger, El coefficient values ​​for the coordinate position (i, j) at which was referred to as Eji, j),
Figure CN102708568AC00051
中' Ca 表不常数; 计算ROTg和Rdis2幅图像中所有坐标位置相同的两个重叠块的结构方向失真映射图,将该结构方向失真映射图的系数矩阵记为Ek,对于Ek中坐标位置为(i,j)处的系数值,将其记为EK(i,j), In 'Ca table is not constant; structural direction and calculating ROTg Rdis2 all images in the same coordinate positions of two overlapping blocks distortion map, the distortion direction of the coefficient matrix structure referred to as a map Ek, Ek for the coordinate position coefficient values ​​(i, j) at which was referred to as EK (i, j),
Figure CN102708568AC00052
⑥计算Lots和Ldis的结构失真评价值,记为 Lots and ⑥ calculated evaluation value Ldis structural distortion, referred to as
Figure CN102708568AC00053
Figure CN102708568AC00054
其中, among them,
Figure CN102708568AC00055
O1表示Lots和Ldis中敏感区域的权重值,W2表示Lots和Ldis中非敏感区域的权重值; 计算ROTg和Rdis的结构失真评价值,记为 O1 denotes a weight value and Ldis Lots sensitive region, W2 represents the weight values ​​Lots and Central Africa Ldis sensitive area; ROTg distortion computing architecture and Rdis evaluation value, referred to as
Figure CN102708568AC00056
其中, « ' !表示R„g和Rdis中敏感区域的权重值,« ' 2表示ROTg和Rdis中非敏感区域的权重值; ⑦根据Q1和QH十算待评价的失真的立体图像Sdis相对于原始的无失真的立体图像Sots的空间频率相似度度量,记为Qf, Qf=^1XQl+(1-^1) XQk,其中,P i表示Ql的权值; ⑧计算Lots和Rots的绝对差值图像,以矩阵形式表示为 Wherein «weighting value 2 represents ROTg and Rdis Central African sensitive areas 'represents R" g and Rdis the weight values ​​sensitive areas, «!'; Stereoscopic image Sdis ⑦ The Q1 and QH ten distortion calculation to be evaluated with respect to undistorted spatial frequency of the original stereoscopic image Sots similarity metric, referred to as Qf, Qf = ^ 1XQl + (1- ^ 1) XQk, where, P i represents the weight of Ql; ⑧ Lots and calculating the absolute difference Rots image, represented as a matrix form
Figure CN102708568AC00057
计算Ldis和Rdis的绝对差值图像,以矩阵形式表示为At,£巧=I 4-4J其中,“II”为取绝对值符号;⑨将和Z/ 2幅图像分别分割成个互不重叠的尺寸大小为8X8的块,然后对D ^和/)= 2幅图像中的所有块分别实施奇异值分解,得到对应的由其每个块的奇异值矩阵组成的奇异值映射图和对应的由其每个块的奇异值矩阵组成的奇异值映射图,将实施奇异值分解后得到的奇异值映射图的系数矩阵记为Gots,对于Gorg中第n个块的奇异值矩阵中坐标位置为(P,q)处的奇异值,将其记为<^1(几?),将1)=实施奇异值分解后得到的奇异值映射图的系数矩阵记为Gdis,对于Gdis中第n个块的奇异值矩阵中坐标位置为(P,q)处的奇异值,将其记为<5:(凡0,其中,Wui表示DJ和£@的宽,Hui表示和Pf/ X Hn* 的高, And Rdis Ldis calculated absolute difference image, expressed in matrix form as At, where £ = I 4-4J Qiao, "II" as the absolute value symbol; ⑨ and the Z / 2 is divided into two images do not overlap each the size of 8X8 blocks, and then D ^ /) = 2 for all image blocks each embodiment singular value decomposition, the singular value map obtained by the composition of each block and the corresponding singular value matrix corresponding Singular value map by singular value matrix consisting of each of the blocks, the singular value map embodiment coefficient matrix referred to the singular value decomposition is obtained Gots, Gorg singular value for n-th block matrix coordinate position Singular value (P, q) at which was referred to as <^ 1 (few?), 1) = embodiments coefficient singular value map the singular value decomposition of the matrix is ​​referred to as Gdis, for Gdis n-th singular value matrix block coordinate position (P, q) at the singular values, which will be referred to as <5 :( 0 where, where, and DJ £ @ Wui represents the width, and Hui represents Pf / X Hn * of high,
Figure CN102708568AC00061
⑩计算DSr对应的奇异值映射图和对应的奇异值映射图的奇异值偏差评价值, Singular value deviation ⑩ DSr evaluation value calculation map corresponding to the singular values ​​and corresponding singular value map,
Figure CN102708568AC00062
其中, among them,
Figure CN102708568AC00063
dgCPJ)表不GOTg中第n个块的奇异值矩阵中坐标位置为(p, p)处的奇异值,Gls(p,p)轰示Gdis中第n个块的奇异值矩阵中坐标位置为(p,p)处的奇异值; ⑪对和< 分别实施奇异值分解,分别得到和Dg各自对应的2个正交矩阵和I个奇异值矩阵,将实施奇异值分解后得到的2个正交矩阵分别记为X _和VOTg,将1¾?实施奇异值分解后得到的奇异值矩阵记为0OTg,XorgxOorgxVorg= D0J,将D=实施奇异值分解后得到的2个正交矩阵分别记为X dis和Vdis,将实施奇异值分解后得到的奇异值矩阵记为Odis, Zdn X°ltn XK/n = Dm ; 12分别计算和I幅图像剥夺奇异值后的残留矩阵图,将/¾?剥夺奇异值后的残留矩阵图记为Xorg, Xorg=XorgX A XVots,将剥夺奇异值后的残留矩阵图记为Xdis,Xdis= X disX A XVdis,其中,A表示单位矩阵,A的大小与0OTg和Odis大小一致; 计算Xorg和Xdis的均值偏差率,记为 Matrix of singular values ​​in the coordinate position dgCPJ) table does not GOTg n-th block of (p, Singular Value p) at, Gls (p, p) H shown singular value Gdis n-th block of the matrix coordinate position Singular value (p, p) at; and ⑪ of <singular value decomposition of each embodiment, respectively, and each corresponding Dg two orthogonal matrices and singular value matrix I, the embodiments of the singular value decomposition to give two positive deposit matrix are denoted as X _ and VOTg, the 1¾? embodiment the singular value decomposition of the matrix of singular values ​​obtained referred to as 0OTg, XorgxOorgxVorg = D0J, the embodiment D = 2 orthogonal matrices obtained after the singular value decomposition are referred to as X dis and Vdis, the embodiment of the singular value matrix referred to the singular value decomposition of Odis, Zdn X ° ltn XK / n = Dm; 12 were calculated and the residual matrix view of the I images deprivation singular values, the / ¾ deprivation? the residue after the stamp matrix singular value Xorg, Xorg = XorgX a XVots, the residue matrix after the deprivation stamp singular value Xdis, Xdis = X disX a XVdis, wherein, a represents a unit matrix, and a is the size of 0OTg Odis same size; and calculate the mean deviation rate Xorg Xdis, denoted
Figure CN102708568AC00064
,其中,X表示,Xorg和Xdis中的像素点的横坐标,y表示Xots和Xdis中的像素点的纵坐标; #计算待评价的失真的立体图像Sdis相对于原始的无失真的立体图像Sots的立体感知255评价度量,记为Qs,G ——,其中,T表示常数,用于调节K和在Qs中所起的A. + r X ¢7 (p重要性; ⑮根据QjP Qs,计算待评价的失真的立体图像Sdis的图像质量评价分值,记为Q,Q=QfX(Qs) P,其中,P表示权重系数值。 Wherein, X represents the abscissa of the pixel in Xorg and Xdis, y represents the vertical coordinate of the pixel in Xots and Xdis; # distortion calculating a stereoscopic image to be evaluated with respect to Sdis original undistorted stereoscopic image Sots stereoscopic perception ratings metrics 255, referred to as Qs, G -, where, T represents a constant, and K for adjusting the play in the Qs A. + r X ¢ 7 (p importance; ⑮ according QjP Qs, calculate image quality evaluation Sdis stereoscopic image distortion value to be evaluated, referred to as Q, Q = QfX (Qs) P, where, P represents the weight coefficient values.
2.根据权利要求I所述的一种基于结构失真的立体图像客观质量评价方法,其特征在于所述的步骤②中Lots和Ldis各自对应的敏感区域矩阵映射图的系数矩阵Al的获取过程为: ②-al、对Lots作水平及垂直方向Sobel算子处理,得到LOTg的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, n和Zv, n,然后计算Lots的梯度幅值图,记为Z11, An I according to claim objective quality assessment method of the stereoscopic image distortion based on the structure, characterized in that the coefficient matrix of the Al matrix sensitive area of ​​the map in the step ② corresponding to each of Lots Ldis and acquisition process is : ②-al, Lots of horizontal and vertical directions as a Sobel operator to give LOTg horizontal gradient image and a vertical direction gradient image, denoted as Zh, n and Zv, n, and then calculate the gradient magnitude Lots FIG. denoted Z11,
Figure CN102708568AC00071
,其中,Z11 (x, y)表示Z11 中坐标位置为(x,y)的像素点的梯度幅值,Zhj n (x, y)表示Zh, u中坐标位置为(x,y)的像素点的水平方向梯度值,Zv, n (x,y)表示Zv, u中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W',I ^ y ^ H/,此处W'表示Z11的宽,H'表示Z11的高; ②_a2、对Ldis作水平及垂直方向Sobel算子处理,得到Ldis的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, 12和Zv, 12,然后计算Ldis的梯度幅值图,记为Z12, Wherein, Z11 (x, y) represents the coordinate position Z11 (x, y) of the pixel gradient magnitude, Zhj n (x, y) represents Zh, u in the coordinate position (x, y) of the pixel horizontal gradient value point, Zv, n (x, y) represents Zv, u in the coordinate position (x, y) in the vertical direction of the gradient values ​​of pixel points, I San x San W ', I ^ y ^ H / where W 'denotes the width of Z11, H' represents the high-Z11; ②_a2, Ldis of horizontal and vertical directions for a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Ldis, denoted as Zh, 12 and Zv, 12, and then calculate the gradient magnitude FIG Ldis, referred to as Z12,
Figure CN102708568AC00072
,其中,Z12 (x, y)表示Z12 中坐标位置为(x, y)的像素点的梯度幅值,Zh,12(x,y)表示Zh,12中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,12(x,y)表示Zv,12中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W',I ^ y ^ H/,此处W'表示Z12的宽,H'表示Z12的高; ②-a 3、 计算划分区域时所需的阈值T, Wherein, Z12 (x, y) represents the coordinate position Z12 (x, y) of the pixel gradient magnitude, Zh, 12 (x, y) represents Zh, 12 the coordinate position (x, y) of horizontal gradient value of the pixel point, Zv, 12 (x, y) represents Zv, 12 the coordinate position (x, y) in the vertical direction of the gradient values ​​of pixel points, I San x San W ', I ^ y ^ H /, where W 'denotes the width of Z12, H' represents the high Z12; ②-a 3, a desired divided area when calculating the threshold T,
Figure CN102708568AC00073
其中,a 为常数,Z11 (x,y)表示Z11 中坐标位置为(X,y)的像素点的梯度幅值,Z12 (X,y)表示Z12中坐标位置为(x, y)的像素点的梯度幅值; ②-a4、将Z11中坐标位置为(i, j)的像素点的梯度幅值记为Z11Q, j),将Z12中坐标位置为(i,j)的像素点的梯度幅值记为Z12 (i, j),判断Z11 (i, j) >T或Z12 (i, j) >T是否成立,如果成立,则确定Lots和Ldis中坐标位置为(i,j)的像素点属于敏感区域,并令\(i,j)=l,否贝U,确定Lots和Ldis中坐标位置为(i,j)的像素点属于非敏感区域,并令\(i,j)=0,其中, Wherein, a is a constant, Z11 (x, y) represents the coordinate position Z11 (X, y) of the pixel gradient magnitude, Z12 (X, y) represents the coordinate position Z12 (x, y) of the pixel gradient amplitude points; ②-a4, the gradient of a pixel in the coordinate position Z11 (i, j) is referred to as amplitude Z11Q, j), the coordinate position Z12 (i, j) of the pixel referred to as gradient magnitude Z12 (i, j), is determined Z11 (i, j)> T or Z12 (i, j)> T is satisfied, if established, is determined, and Lots Ldis coordinate position (i, j) the pixels belonging to the sensitive areas, and so \ (i, j) = l, no shellfish U, determined Lots and Ldis coordinate position (i, j) of pixels belonging to the non-sensitive region, and so \ (i, j ) = 0, wherein
Figure CN102708568AC00074
所述的步骤②中Rots和Rdis各自对应的敏感区域矩阵映射图的系数矩阵Ak的获取过程为: ②-bl、对Rots作水平及垂直方向Sobel算子处理,得到Rots的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, rl和Zv, rl,然后计算Rots的梯度幅值图,记为Zrt, Acquisition process coefficient matrix Ak matrix sensitive area of ​​the map in the step ② and Rdis Rots respectively corresponding to: ②-bl, Rots of horizontal and vertical directions as a Sobel operator to give the horizontal gradient image and Rots vertical gradient images, respectively referred to as a Zh, rl, and Zv, rl, and then calculate the gradient magnitude FIG Rots, denoted Zrt,
Figure CN102708568AC00075
,其中,Zrl (x, y)表示Zrl 中坐标位置为(x, y)的像素点的梯度幅值,Zhj rl (x, y)表示Zh, 中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,rt(x,y)表示Zv,ri中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W',1 ^ y ^ H/,此处W'表示Zrt的宽,H'表示Zrt的高; ②_b2、对Rdis作水平及垂直方向Sobel算子处理,得到Rdis的水平方向梯度图像和垂直方向梯度图像,分别记为Zh, r2和Zv,r2,然后计算Rdis的梯度幅值图,记为L,Zr2(X,y) 二y](Zh r2(x,y)f + (Zvr2(x,y)f,其中,Zr2 (x, y)表示Zr2 中坐标位置为(x, y)的像素点的梯度幅值,Zhj r2(x, y)表示Zh, r2中坐标位置为(x,y)的像素点的水平方向梯度值,Zv,r2(x,y)表示Zv,r2中坐标位置为(x,y)的像素点的垂直方向梯度值,I彡x彡W',I ≤ y ≤ H/,此处W'表示Zrf的宽,H'表示Zrt的高;②-b3、计算划分区域时所需的阈值T ', Wherein, Zrl (x, y) represents a coordinate ZRL position (x, y) of the pixel gradient magnitude, Zhj rl (x, y) represents Zh, the coordinate position (x, y) of the pixel the horizontal gradient value, Zv, rt (x, y) represents Zv, ri the coordinate position (x, y) in the vertical direction of the gradient values ​​of pixel points, I San x San W ', 1 ^ y ^ H /, where W 'denotes the width Zrt, H' represents the high Zrt; ②_b2, Rdis of horizontal and vertical directions for a Sobel operator to give the horizontal direction and the vertical direction gradient image of the gradient image Rdis, denoted as Zh, r2 and Zv, r2, and then calculate the gradient magnitude FIG Rdis, denoted by L, Zr2 (X, y) two y] (Zh r2 (x, y) f + (Zvr2 (x, y) f, wherein, Zr2 (x , y) represents Zr2 coordinate position (X, y) of the pixel gradient magnitude, Zhj r2 (x, y) represents Zh, r2 the coordinate position (X, y) in the horizontal direction of the gradient of the pixel values , Zv, r2 (x, y) represents Zv, r2 the coordinate position (x, y) of pixels in the vertical direction of the gradient value, I x San San W ', I ≤ y ≤ H /, where W' denotes Zrf wide, H 'represents the high Zrt; ②-b3, required for calculating the threshold value divided area T',
Figure CN102708568AC00081
其中,a 为常数,Zrl(x,y)表示Zrl 中坐标位置为(X,y)的像素点的梯度幅值,Zr2 (X,y)表示Zr2中坐标位置为(x,y)的像素点的梯度幅值; ②_b4、将Zrt中坐标位置为(i, j)的像素点的梯度幅值记为Zrt (i, j),将Zr2中坐标位置为(i,j)的像素点的梯度幅值记为Zrf (i, j),判断rLxx (i, j) >T或Zrt (i, j) >T是否成立,如果成立,则确定Rots和Rdis中坐标位置为(i,j)的像素点属于敏感区域,并令AK(i,j)=l,否贝U,确定&8和Rdis中坐标位置为(i,j)的像素点属于非敏感区域,并令AK(i,j)=0,其中,0≤ i ≤(W-8), 0≤ j ≤ (H-8)。 Wherein, a is a constant, Zrl (x, y) represents a coordinate position in ZRL (X, y) of the pixel gradient magnitude, Zr2 (X, y) represents the coordinate position Zr2 (x, y) of the pixel gradient amplitude points; ②_b4, the gradient of a pixel in Zrt coordinate position (i, j) is referred to as amplitude Zrt (i, j), the coordinate position Zr2 (i, j) of the pixel referred to as gradient magnitude Zrf (i, j), it is determined rLxx (i, j)> T or Zrt (i, j)> T is satisfied, if established, is determined and Rdis Rots coordinate position (i, j) the pixels belonging to the sensitive areas, and so AK (i, j) = l, no shellfish U, determined & 8 and Rdis coordinate position (i, j) of pixels belonging to the non-sensitive region, and so AK (i, j ) = 0, wherein, 0≤ i ≤ (W-8), 0≤ j ≤ (H-8).
3.根据权利要求I或2所述的一种基于结构失真的立体图像客观质量评价方法,其特征在于所述的步骤⑦中P !的获取过程为: ⑦-I、采用n幅无失真的立体图像建立其在不同失真类型不同失真程度下的失真立体图像集,该失真立体图像集包括多幅失真的立体图像,其中,n > I ; ⑦-2、利用主观质量评价方法获取失真立体图像集中的每幅失真的立体图像的平均主观评分差值,记为DMOS,DM0S=100-M0S,其中,MOS表示主观评分均值,DMOS G [0,100]; ⑦-3、根据步骤①至步骤⑥的操作过程,计算失真立体图像集中的每幅失真的立体图像的左视点图像相对于对应的无失真的立体图像的左视点图像的敏感区域的评价值Qw和非敏感区域的评价值Qm^ ; ⑦-4、采用数学拟合方法拟合失真立体图像集中失真的立体图像的平均主观评分差值DMOS和对应的Qm, L和Qnm, L,从而获得P i值。 The one or more I according to claim 2 objective quality assessment method of the stereoscopic image distortion based on the structure, wherein said step of P ⑦ acquisition process is:! ⑦-I, using n as undistorted the stereoscopic image set to establish its distortion stereoscopic image distortion at different levels of the different types of distortion, the distortion of a stereoscopic image set comprises a plurality of stereoscopic image distortion, wherein, n> I; ⑦-2, using the subjective quality assessment method of obtaining a stereoscopic image distortion mean opinion score difference of each piece of stereoscopic image distortion set, referred to as DMOS, DM0S = 100-M0S, wherein, the MOS represents the mean subjective score, DMOS G [0,100]; ⑦-3, ① to step according to step during operation ⑥ calculated distortion left-view image of the stereoscopic image of each piece of distorted perspective image set with respect to the evaluation value of the evaluation value Qw and non-sensitive regions sensitive area left viewpoint image distortion-free stereoscopic image corresponding to Qm ^ ; ⑦-4, using mathematical fitting method of fitting a stereoscopic image distortion stereoscopic image concentration difference between the mean opinion score distortion and corresponding DMOS Qm, L and Qnm, L, to thereby obtain P i value.
CN 201210145034 2012-05-11 2012-05-11 Stereoscopic image objective quality evaluation method on basis of structural distortion CN102708568B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201210145034 CN102708568B (en) 2012-05-11 2012-05-11 Stereoscopic image objective quality evaluation method on basis of structural distortion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201210145034 CN102708568B (en) 2012-05-11 2012-05-11 Stereoscopic image objective quality evaluation method on basis of structural distortion

Publications (2)

Publication Number Publication Date
CN102708568A true CN102708568A (en) 2012-10-03
CN102708568B CN102708568B (en) 2014-11-05

Family

ID=46901288

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201210145034 CN102708568B (en) 2012-05-11 2012-05-11 Stereoscopic image objective quality evaluation method on basis of structural distortion

Country Status (1)

Country Link
CN (1) CN102708568B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103108209A (en) * 2012-12-28 2013-05-15 宁波大学 Stereo image objective quality evaluation method based on integration of visual threshold value and passage
CN104574363A (en) * 2014-12-12 2015-04-29 南京邮电大学 Full reference image quality assessment method in consideration of gradient direction difference
CN104036502B (en) * 2014-06-03 2016-08-24 宁波大学 A non-stereoscopic image distortion reference fuzzy quality evaluation method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050089246A1 (en) * 2003-10-27 2005-04-28 Huitao Luo Assessing image quality
US20080226147A1 (en) * 2007-03-16 2008-09-18 Sti Medical Systems, Llc Method to provide automated quality feedback to imaging devices to achieve standardized imaging data
US20080232667A1 (en) * 2007-03-22 2008-09-25 Fujifilm Corporation Device, method and recording medium containing program for separating image component, and device, method and recording medium containing program for generating normal image
CN101833766A (en) * 2010-05-11 2010-09-15 天津大学 Stereo image objective quality evaluation algorithm based on GSSIM
CN101872479A (en) * 2010-06-09 2010-10-27 宁波大学 Three-dimensional image objective quality evaluation method
CN102142145A (en) * 2011-03-22 2011-08-03 宁波大学 Image quality objective evaluation method based on human eye visual characteristics

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050089246A1 (en) * 2003-10-27 2005-04-28 Huitao Luo Assessing image quality
US20080226147A1 (en) * 2007-03-16 2008-09-18 Sti Medical Systems, Llc Method to provide automated quality feedback to imaging devices to achieve standardized imaging data
US20080232667A1 (en) * 2007-03-22 2008-09-25 Fujifilm Corporation Device, method and recording medium containing program for separating image component, and device, method and recording medium containing program for generating normal image
CN101833766A (en) * 2010-05-11 2010-09-15 天津大学 Stereo image objective quality evaluation algorithm based on GSSIM
CN101872479A (en) * 2010-06-09 2010-10-27 宁波大学 Three-dimensional image objective quality evaluation method
CN102142145A (en) * 2011-03-22 2011-08-03 宁波大学 Image quality objective evaluation method based on human eye visual characteristics

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
周俊明等: "利用奇异值分解法的立体图像客观质量评价模型", 《计算机辅助设计与图形学学报》, vol. 23, no. 5, 31 May 2011 (2011-05-31), pages 870 - 877 *
沈丽丽等: "基于三维特征和结构相似度的图像质量评价方法", 《光电子·激光》, vol. 21, no. 11, 30 November 2010 (2010-11-30), pages 1713 - 1719 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103108209A (en) * 2012-12-28 2013-05-15 宁波大学 Stereo image objective quality evaluation method based on integration of visual threshold value and passage
CN103108209B (en) * 2012-12-28 2015-03-11 宁波大学 Stereo image objective quality evaluation method based on integration of visual threshold value and passage
CN104036502B (en) * 2014-06-03 2016-08-24 宁波大学 A non-stereoscopic image distortion reference fuzzy quality evaluation method
CN104574363A (en) * 2014-12-12 2015-04-29 南京邮电大学 Full reference image quality assessment method in consideration of gradient direction difference
CN104574363B (en) * 2014-12-12 2017-09-29 南京邮电大学 An all reference gradient direction image quality evaluation method considering the difference

Also Published As

Publication number Publication date
CN102708568B (en) 2014-11-05

Similar Documents

Publication Publication Date Title
Sazzad et al. Stereoscopic image quality prediction
Wang et al. Quality prediction of asymmetrically distorted stereoscopic 3D images
Shao et al. Full-reference quality assessment of stereoscopic images by learning binocular receptive field properties
JP6250916B2 (en) 3-dimensional video quality prediction system of
CN102682446A (en) Method and apparatus for generating a dense depth map using an adaptive joint bilateral filter
CN102170581B (en) Human-visual-system (HVS)-based structural similarity (SSIM) and characteristic matching three-dimensional image quality evaluation method
CN102333233B (en) Stereo image quality objective evaluation method based on visual perception
CN102209257B (en) Stereo image quality objective evaluation method
CN102902961B (en) Face super-resolution processing method based on K neighbor sparse coding average value constraint
CN101872479B (en) Three-dimensional image objective quality evaluation method
CN102750695B (en) Machine learning-based stereoscopic image quality objective assessment method
CN101572828B (en) Method for correcting distortion of camera and video camera in real time based on GPU
CN101877143A (en) Three-dimensional scene reconstruction method of two-dimensional image group
CN101562758A (en) Method for objectively evaluating image quality based on region weight and visual characteristics of human eyes
CN101378519B (en) Method for evaluating quality-lose referrence image quality base on Contourlet transformation
CN101610425B (en) Method for evaluating stereo image quality and device
CN102075786B (en) Method for objectively evaluating image quality
CN103873741A (en) Method and device for substituting area of interest in video
CN102142145B (en) Image quality objective evaluation method based on human eye visual characteristics
Wang et al. Novel spatio-temporal structural information based video quality metric
CN101726855A (en) Correction method of fisheye image distortion on basis of cubic projection
CN101872472B (en) Method for super-resolution reconstruction of facial image on basis of sample learning
CN102137271A (en) Method and device for evaluating image quality
Li et al. Visual discomfort of stereoscopic 3D videos: Influence of 3D motion
CN102421007A (en) Image quality evaluating method based on multi-scale structure similarity weighted aggregate

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C14 Grant of patent or utility model