Summary of the invention
Technical problem to be solved by this invention provides a kind of maximum changing range that the quality of the left relatively visual point image of quality of right visual point image can descend can determine asymmetric stereo scopic video coding well the time, reach the purpose that improves the coding compression efficiency by the quality that reduces right visual point image, utilize the stereoscopic vision masking effect to make the observer can not perceive the decline of right view-point image quality simultaneously, can effectively guarantee the objective analysis method of the minimum discernable change step of human eye of the total quality of stereo-picture.
The present invention solves the problems of the technologies described above the technical scheme that adopts: a kind of objective analysis method of minimum discernable change step of stereo-picture is characterized in that may further comprise the steps:
1. make S
OrgUndistorted stereo-picture for original makes S
DisFor the stereo-picture of distortion to be evaluated, with S
OrgLeft visual point image be designated as L
Org, with S
OrgRight visual point image be designated as R
Org, with S
DisLeft visual point image be designated as L
Dis, with S
DisRight visual point image be designated as R
Dis
2. to L
Org, R
Org, L
DisAnd
R
Dis4 width of cloth images are implemented singular value decomposition respectively, obtain L respectively
Org, R
Org, L
DisAnd R
DisEach self-corresponding singular value vector of 4 width of cloth images is with L
OrgThe singular value vector be designated as
With R
OrgThe singular value vector be designated as
With L
DisThe singular value vector be designated as
With R
DisThe singular value vector be designated as
Wherein, the dimension of each singular value vector is m, and m=min (M, N), min () is for getting minimum value function, the horizontal size size of M presentation video, the vertical dimension size of N presentation video;
3. calculate L
OrgThe singular value vector
With L
DisThe singular value vector
The absolute difference vector, be designated as X
L,
With X
LAs L
DisCharacteristic vector, calculate R
OrgThe singular value vector
With R
DisThe singular value vector
The absolute difference vector, be designated as X
R,
With X
RAs R
DisCharacteristic vector, wherein, " || " is the symbol that takes absolute value;
4. to L
DisCharacteristic vector X
LAnd R
DisCharacteristic vector X
RCarry out linear weighted function, obtain S
DisCharacteristic vector, be designated as X, X=w
LX
L+ w
RX
R, wherein, w
LExpression L
DisWeights proportion, w
RExpression R
DisWeights proportion, w
L+ w
R=1;
5. adopt n undistorted stereo-picture, set up its distortion stereo-picture set under the different distortion levels of coding distortion type H.264, this distortion stereo-picture set comprises the stereo-picture of several distortions, utilizes the subjective quality evaluation method to obtain the average subjective scoring difference of the stereo-picture of every width of cloth distortion in the set of distortion stereo-picture respectively, is designated as DMOS, DMOS=100-MOS, wherein, MOS represents the subjective scoring average, DMOS ∈ [0,100], n 〉=1;
6. adopt and calculating S
DisThe identical method of characteristic vector X, the different weights proportion combined feature vectors of the stereo-picture of every width of cloth distortion during the calculated distortion stereo-picture is gathered respectively, j kind weights proportion combined feature vector for the stereo-picture of i width of cloth distortion in the set of distortion stereo-picture is designated as X with it
I, j, wherein, 1≤i≤n', 1≤j≤m', n' represent the width of cloth number of the stereo-picture of the distortion that comprises in the distortion stereo-picture set, m' represents the kind number of all weights proportions combinations;
7. adopt support vector regression that the identical weights proportion combined feature vector of the stereo-picture of distortions all in the set of distortion stereo-picture is trained, obtain different weights than the support vector regression training pattern of recombination, for the support vector regression training pattern of j kind weights than recombination, it is designated as f
j(X
Inp), wherein, f
j() is that j kind weights are than the function representation form of the regression function of recombination, X
InpExpress support for the input vector of vector regression training pattern;
8. from n undistorted stereo-picture, choose a width of cloth stereo-picture arbitrarily, appoint then and get a coded quantization parameter the left visual point image of this stereo-picture is encoded, and adopt a plurality of different coded quantization parameters that the right visual point image of this stereo-picture is encoded, utilize support vector regression training pattern that training obtains that several test patterns that the width of cloth left side visual point image that obtained by coding and several right visual point images constitute are tested, but calculate the minimum discernable change step of human eye perception stereoscopic vision right visual point image in the test pattern when changing.
Described step detailed process 2. is:
2.-1, with size be the L of M * N
OrgBe expressed as the two-dimensional matrix of M * N dimension, be designated as
By the two-dimensional matrix of singular value decomposition with M * N dimension
Be expressed as
Wherein,
The orthogonal matrix of expression M * M dimension,
The orthogonal matrix of expression N * N dimension,
Expression
Transposed matrix,
The diagonal matrix of expression M * N dimension;
2.-2, the diagonal matrix that M * N is tieed up
Diagonal element as the two-dimensional matrix of M * N dimension
Singular value, from the two-dimensional matrix of M * N dimension
Singular value in take out the singular value formation L of m non-zero
OrgThe singular value vector, be designated as
Wherein, and m=min (M, N), min () is for getting minimum value function;
2.-3, to R
Org, L
DisAnd R
DisAdopt with step 2.-1 to 2.-2 identical operations, obtain R
Org, L
DisAnd R
DisThe singular value vector, be designated as respectively
With
Described step detailed process 7. is:
7.-1, with the identical weights proportion combined feature vector of the stereo-picture of distortions all in the distortion stereo-picture set and average subjective scoring difference as the set of training sample data, j kind weights are designated as Ω than the training sample data set of recombination
Q, j, { X
K, j, DMOS
k∈ Ω
Q, j, wherein, 1≤j≤m', m' represent the kind number of all weights proportion combinations, q represents that j kind weights are than the training sample data set omega of recombination
Q, jIn the width of cloth number of stereo-picture of the distortion that comprises, X
K, jRepresent that j kind weights are than the training sample data set omega of recombination
Q, jIn the characteristic vector of stereo-picture of k width of cloth distortion, DMOS
kRepresent that j kind weights are than the training sample data set omega of recombination
Q, jIn the average subjective scoring difference of stereo-picture of k width of cloth distortion, 1≤k≤q;
7.-2, structure X
K, jRegression function f
j(X
K, j),
Wherein, f
j() be j kind weights than the function representation form of the regression function of recombination, w is weight vector, w
TBe the transposed matrix of w, b is bias term,
Represent that j kind weights are than the training sample data set omega of recombination
Q, jIn the characteristic vector X of stereo-picture of k width of cloth distortion
K, jLinear function,
Be the kernel function in the support vector regression,
X
L, jBe that j kind weights are than the training sample data set omega of recombination
Q, jIn the characteristic vector of stereo-picture of l width of cloth distortion, γ is nuclear parameter, is used for the scope of reflection input sample value, the scope of sample value is more big, and the γ value is also just more big, and exp () expression is the exponential function at the end with e, e=2.71828183, " || || " for asking the Euclidean distance symbol;
7.-3, adopt support vector regression to the training sample data set omega of j kind weights than recombination
Q, jIn the characteristic vector of stereo-picture of all distortion train, make error minimum between the regression function value that obtains through training and the average subjective scoring difference, match obtains the weight vector w of optimum
OptBias term b with optimum
Opt, with the weight vector w of optimum
OptBias term b with optimum
OptCombination be designated as (w
Opt, b
Opt),
Utilize optimum weight vector w
OptBias term b with optimum
OptConstruct j kind weights than the support vector regression training pattern of recombination, be designated as f
j(X
Inp),
Wherein, Ψ represents the training sample data set omega of j kind weights than recombination
Q, jIn the set of combination of the characteristic vector of stereo-picture of all distortion all weight vector of training and bias term,
Expression minimizes probability density function, X
InpExpress support for the input vector of vector regression training pattern, (w
Opt)
TBe w
OptTransposed matrix,
Express support for the input vector X of vector regression training pattern
InpLinear function.
Described step detailed process 8. is:
8.-1, from n undistorted stereo-picture, choose a width of cloth stereo-picture arbitrarily, appoint and get a coded quantization parameter as the basic coding quantization parameter of the left visual point image of this stereo-picture, be designated as QP
L, will adopt QP
LThe encode left visual point image that obtains of the left visual point image of this stereo-picture is defined as the left visual point image of A quality point, is designated as L
A
8.-2, obtain five values more than or equal to QP
LAnd the coded quantization parameter that value has nothing in common with each other is as the basic coding quantization parameter of the right visual point image of this stereo-picture, adopt these five basic coding quantization parameters respectively the right visual point image of this stereo-picture to be encoded, obtain the right visual point image that five width of cloth quality have nothing in common with each other, be defined as the right visual point image of A quality point, the right visual point image of B quality point, the right visual point image of C quality point, the right visual point image of D quality point, the right visual point image of E quality point respectively, be designated as R respectively
A, R
B, R
C, R
D, R
E, wherein, with the right visual point image R of A quality point
AThe basic coding quantization parameter that adopts during coding is designated as QP
RA, with the right visual point image R of B quality point
BThe basic coding quantization parameter that adopts during coding is designated as QP
RB, with the right visual point image R of C quality point
CThe basic coding quantization parameter that adopts during coding is designated as QP
RC, with the right visual point image R of D quality point
DThe basic coding quantization parameter that adopts during coding is designated as QP
RD, with the right visual point image R of E quality point
EThe basic coding quantization parameter that adopts during coding is designated as QP
RE
8.-3, with the left visual point image L of A quality point
ARight visual point image R with the A quality point
AThe stereo-picture that constitutes is as the 1st width of cloth test pattern, and is designated as S
AALeft visual point image L with the A quality point
ARight visual point image R with the B quality point
BThe stereo-picture that constitutes is as the 2nd width of cloth test pattern, and is designated as S
ABLeft visual point image L with the A quality point
ARight visual point image R with the C quality point
CThe stereo-picture that constitutes is as the 3rd width of cloth test pattern, and is designated as S
ACLeft visual point image L with the A quality point
ARight visual point image R with the D quality point
DThe stereo-picture that constitutes is as the 4th width of cloth test pattern, and is designated as S
ADLeft visual point image L with the A quality point
ARight visual point image R with the E quality point
EThe stereo-picture that constitutes is as the 5th width of cloth test pattern, and is designated as S
AE
8.-4, adopt and calculate S
DisThe identical method of characteristic vector X, calculate S
AAAdopt different weights proportion combined feature vectors, with S
AAAdopt j kind weights proportion combined feature vector to be designated as X
A, j, wherein, 1≤j≤m', m' represent the kind number of all weights proportion combinations;
8.-5, according to the support vector regression training pattern, the prediction S
AAAdopt the evaluating objective quality predicted value of all different weights proportion combined feature vectors, for S
AAAdopt j kind weights proportion combined feature vector X
A, j, with X
A, jAs the input vector of j kind weights than the support vector regression training pattern of recombination, prediction obtains X
A, jThe evaluating objective quality predicted value, be designated as Q
j, Q
j=f
j(X
A, j),
Wherein, w
OptBe the weight vector of optimum, (w
Opt)
TBe w
OptTransposed matrix, b
OptBe the bias term of optimum,
Expression X
A, jLinear function;
8.-6, with S
AAAdopt the mean value of evaluating objective quality predicted value of all different weights proportion combined feature vectors as S
AAThe objective evaluation value, be designated as Score
A,
8.-7,8.-4 employing and step to 8.-6 identical operations, obtain S
AB, S
AC, S
ADAnd S
AEThe objective evaluation value, be designated as Score respectively
B, Score
C, Score
DAnd Score
E
8.-8, calculate S
AA, S
AB, S
AC, S
ADAnd S
AEObjective score value, be designated as D respectively
1, D
2, D
3, D
4And D
5, D
1=0,
8.-9, from { D
J'| the maximum objective score value of the value of finding out among 1≤j'≤5} is designated as D
Max, with D
MaxThe quality point of corresponding right visual point image is designated as q, the stereo-picture that is constituted by the left visual point image of the right visual point image of q quality point and A quality point then, and the picture quality difference that comes with the English alphabet order between the stereo-picture of the corresponding right visual point image of each quality point and the left visual point image formation of A quality point before the q quality point is distinguishable, the difference of the Y-PSNR of the Y-PSNR of the right visual point image of definition q quality point and the right visual point image of A quality point is as adopting QP
LHuman eye was to the discernable change step of appreciable minimum of asymmetric stereo image coding when left visual point image was encoded.
Described step 6. in the process of the different weights proportion combined feature vectors of the stereo-picture of calculated distortion, j kind weights are designated as (w than recombination
L', w
R'), w
R'=w
0+ (j-1) * and Δ w, w
L'=1-w
R', wherein, w
L' the weights proportion of left visual point image of stereo-picture of expression distortion, w
R' the weights proportion of right visual point image of stereo-picture of expression distortion, w
0=0.55, Δ w=0.05,1≤j≤m', m' represent the kind number of all weights proportion combinations.
Compared with prior art, the invention has the advantages that:
1) the inventive method is mapped in the high-dimensional feature space by the characteristic vector of support vector regression with stereo-picture, in high-dimensional feature space, carry out Linear Estimation again, the characteristic vector of structure optimum regression function stereoscopic image is tested, avoided human visual system's correlation properties and the complicated simulation process of mechanism, and because training sample and test sample book are separate, can avoid test result to the depending on unduly of training data like this, the maximum changing range that the quality of the left relatively visual point image of quality of right visual point image can descend in the time of determining asymmetric stereo scopic video coding better.
2) the inventive method adopts different weights proportion combinations to train by the stereo-picture to the every width of cloth distortion in the set of distortion stereo-picture, obtain different weights than the support vector regression training pattern of recombination, the support vector regression model that utilizes training to obtain is tested any test pattern, under the changeless situation of left view-point image quality, but the critical mass point of right view-point image quality when measuring the variation of human eye perception stereo image quality, thereby determine the maximum changing range that the left relatively view-point image quality of right view-point image quality can descend when stereoscopic image was encoded, can reach the purpose that improves code efficiency by reducing right view-point image quality, can utilize the stereoscopic vision masking effect to make the observer can not perceive the fact that right view-point image quality descends again, thereby guarantee the total quality of stereo-picture.
Embodiment
Describe in further detail below in conjunction with the present invention of accompanying drawing embodiment.
The objective analysis method of the minimum discernable change step of a kind of stereo-picture that the present invention proposes, it totally realizes block diagram as shown in Figure 1, it mainly may further comprise the steps:
1. make S
OrgUndistorted stereo-picture for original makes S
DisFor the stereo-picture of distortion to be evaluated, with S
OrgLeft visual point image be designated as L
Org, with S
OrgRight visual point image be designated as R
Org, with S
DisLeft visual point image be designated as L
Dis, with S
DisRight visual point image be designated as R
Dis
2. to L
Org, R
Org, L
DisAnd
R
Dis4 width of cloth images are implemented singular value decomposition respectively, obtain L respectively
Org, R
Org, L
DisAnd R
DisEach self-corresponding singular value vector of 4 width of cloth images is with L
OrgThe singular value vector be designated as
With R
OrgThe singular value vector be designated as
With L
DisThe singular value vector be designated as
With R
DisThe singular value vector be designated as
Wherein, the dimension of each singular value vector is m, and (M, N), min () is for getting minimum value function, the horizontal size size of M presentation video, the vertical dimension size of N presentation video for m=min.
In the present embodiment, step detailed process 2. is:
2.-1, with size be the L of M * N
OrgBe expressed as the two-dimensional matrix of M * N dimension, be designated as
By the two-dimensional matrix of singular value decomposition with M * N dimension
Be expressed as
Wherein,
The orthogonal matrix of expression M * M dimension,
The orthogonal matrix of expression N * N dimension,
Expression
Transposed matrix,
The diagonal matrix of expression M * N dimension.
2.-2, the diagonal matrix that M * N is tieed up
Diagonal element as the two-dimensional matrix of M * N dimension
Singular value, from the two-dimensional matrix of M * N dimension
Singular value in take out the singular value formation L of m non-zero
OrgThe singular value vector, be designated as
Wherein, (M, N), min () is for getting minimum value function for m=min.
2.-3, to R
Org, L
DisAnd R
DisAdopt with step 2.-1 to 2.-2 identical operations, obtain R
Org, L
DisAnd R
DisThe singular value vector, be designated as respectively
With
4. to L
DisCharacteristic vector X
LAnd R
DisCharacteristic vector X
RCarry out linear weighted function, obtain S
DisCharacteristic vector, be designated as X, X=w
LX
L+ w
RX
R, wherein, w
LExpression L
DisWeights proportion, w
RExpression R
DisWeights proportion, w
L+ w
R=1.
5. adopt n undistorted stereo-picture, set up its distortion stereo-picture set under the different distortion levels of coding distortion type H.264, this distortion stereo-picture set comprises the stereo-picture of several distortions, utilizes the subjective quality evaluation method to obtain the average subjective scoring difference of the stereo-picture of every width of cloth distortion in the set of distortion stereo-picture respectively, is designated as DMOS, DMOS=100-MOS, wherein, MOS represents the subjective scoring average, DMOS ∈ [0,100], n 〉=1.
In the present embodiment, because the stereo-picture of test obtains by H.264 encoding, therefore the type of distortion of training sample and test sample book should be consistent in support vector regression, utilize the stereo-picture as Fig. 2 a and Fig. 2 b formation, the stereo-picture that Fig. 3 a and Fig. 3 b constitute, the stereo-picture that Fig. 4 a and Fig. 4 b constitute, the stereo-picture that Fig. 5 a and Fig. 5 b constitute, the stereo-picture that Fig. 6 a and Fig. 6 b constitute, the stereo-picture that Fig. 7 a and Fig. 7 b constitute, the stereo-picture that Fig. 8 a and Fig. 8 b constitute, the stereo-picture that Fig. 9 a and Fig. 9 b constitute, the stereo-picture that Figure 10 a and Figure 10 b constitute, the stereo-picture that Figure 11 a and Figure 11 b constitute, the stereo-picture that Figure 12 a and Figure 12 b constitute, the stereo-picture that Figure 13 a and Figure 13 b the constitute undistorted stereo-picture of totally 12 width of cloth (n=12) has been set up its distortion stereo-picture set under the different distortion levels of coding distortion type H.264, and the stereo-picture of distortion has 72 width of cloth in this distortion stereo-picture set.
6. adopt and calculating S
DisThe identical method of characteristic vector X, the different weights proportion combined feature vectors of the stereo-picture of every width of cloth distortion during the calculated distortion stereo-picture is gathered respectively, j kind weights proportion combined feature vector for the stereo-picture of i width of cloth distortion in the set of distortion stereo-picture is designated as X with it
I, j, wherein, 1≤i≤n', 1≤j≤m', n' represent the width of cloth number of the stereo-picture of the distortion that comprises in the distortion stereo-picture set, m' represents the kind number of all weights proportions combinations.
In this specific embodiment, because the quality of right visual point image changes in the test pattern, and the stereoscopic vision masking effect of the stereo-picture that human eye constitutes the left visual point image by the right visual point image of different quality and equal in quality is inconsistent, in order can be better the stereoscopic vision masking effect of human eye to be described, the inventive method adopts different weights proportion that the left visual point image of the stereo-picture of every width of cloth distortion and the characteristic vector of right visual point image in the set of distortion stereo-picture are carried out linear weighted function.The inventive method is designated as (w with j kind weights than recombination in the process of the different weights proportion combined feature vectors of the stereo-picture of calculated distortion
L', w
R'), w
R'=w
0+ (j-1) * and Δ w, w
L'=1-w
R', wherein, w
L' the weights proportion of left visual point image of stereo-picture of expression distortion, w
R' the weights proportion of right visual point image of stereo-picture of expression distortion, w
0=0.55, Δ w=0.05.
At this, get m'=9.
7. because the characteristic vector of the stereo-picture of distortion is the higher dimensional space vector, need in higher dimensional space, construct linear decision function and realize non-linear decision function in the former space, (Support Vector Regression SVR) is the method for the non-linear higher dimensional space conversion of a kind of reasonable realization to support vector regression.Therefore the inventive method adopts support vector regression that the identical weights proportion combined feature vector of the stereo-picture of distortions all in the set of distortion stereo-picture is trained, obtain different weights than the support vector regression training pattern of recombination, for the support vector regression training pattern of j kind weights than recombination, it is designated as f
j(X
Inp), wherein, f
j() is that j kind weights are than the function representation form of the regression function of recombination, X
InpExpress support for the input vector of vector regression training pattern.
In the present embodiment, step detailed process 7. is:
7.-1, with the identical weights proportion combined feature vector of the stereo-picture of distortions all in the distortion stereo-picture set and average subjective scoring difference as the set of training sample data, j kind weights are designated as Ω than the training sample data set of recombination
Q, j, { X
K, j, DMOS
k∈ Ω
Q, j, wherein, 1≤j≤m', m' represent the kind number of all weights proportion combinations, q represents that j kind weights are than the training sample data set omega of recombination
Q, jIn the width of cloth number of stereo-picture of the distortion that comprises, X
K, jRepresent that j kind weights are than the training sample data set omega of recombination
Q, jIn the characteristic vector of stereo-picture of k width of cloth distortion, DMOS
kRepresent that j kind weights are than the training sample data set omega of recombination
Q, jIn the average subjective scoring difference of stereo-picture of k width of cloth distortion, 1≤k≤q.
7.-2, structure X
K, jRegression function f
j(X
K, j),
Wherein, f
j() be j kind weights than the function representation form of the regression function of recombination, w is weight vector, w
TBe the transposed matrix of w, b is bias term,
Represent that j kind weights are than the training sample data set omega of recombination
Q, jIn the characteristic vector X of stereo-picture of k width of cloth distortion
K, jLinear function,
Be the kernel function in the support vector regression,
X
L, jBe that j kind weights are than the training sample data set omega of recombination
Q, jIn the characteristic vector of stereo-picture of l width of cloth distortion, γ is nuclear parameter, is used for the scope of reflection input sample value, the scope of sample value is more big, and the γ value is also just more big, and exp () expression is the exponential function at the end with e, e=2.71828183, " || || " for asking the Euclidean distance symbol.
7.-3, adopt support vector regression to the training sample data set omega of j kind weights than recombination
Q, jIn the characteristic vector of stereo-picture of all distortion train, make error minimum between the regression function value that obtains through training and the average subjective scoring difference, match obtains the weight vector w of optimum
OptBias term b with optimum
Opt, with the weight vector w of optimum
OptBias term b with optimum
OptCombination be designated as (w
Opt, b
Opt),
Utilize optimum weight vector w
OptBias term b with optimum
OptConstruct j kind weights than the support vector regression training pattern of recombination, be designated as f
j(X
Inp),
Wherein, Ψ represents the training sample data set omega of j kind weights than recombination
Q, jIn the set of combination of the characteristic vector of stereo-picture of all distortion all weight vector of training and bias term,
Expression minimizes probability density function, X
InpExpress support for the input vector of vector regression training pattern, (w
Opt)
TBe w
OptTransposed matrix,
Express support for the input vector X of vector regression training pattern
InpLinear function.
8. gather the support vector regression training pattern that obtains by training by the distortion stereo-picture, can reflect that the mass change of left visual point image and right visual point image is to the influence of the stereoscopic vision masking effect of human eye, therefore, can utilize this model that the quality of any stereo-picture is measured.From n undistorted stereo-picture, choose a width of cloth stereo-picture arbitrarily, appoint then and get a coded quantization parameter the left visual point image of this stereo-picture is encoded, and adopt a plurality of different coded quantization parameters that the right visual point image of this stereo-picture is encoded, utilize support vector regression training pattern that training obtains that several test patterns that the width of cloth left side visual point image that obtained by coding and several right visual point images constitute are tested, but calculate the minimum discernable change step of human eye perception stereoscopic vision right visual point image in the test pattern when changing.
In the present embodiment, step detailed process 8. is:
8.-1, from n undistorted stereo-picture, choose a width of cloth stereo-picture arbitrarily, appoint and get a coded quantization parameter as the basic coding quantization parameter of the left visual point image of this stereo-picture, be designated as QP
L, will adopt QP
LThe encode left visual point image that obtains of the left visual point image of this stereo-picture is defined as the left visual point image of A quality point, is designated as L
A
8.-2, obtain five values more than or equal to QP
LAnd the coded quantization parameter that value has nothing in common with each other is as the basic coding quantization parameter of the right visual point image of this stereo-picture, adopt these five basic coding quantization parameters respectively the right visual point image of this stereo-picture to be encoded, obtain the right visual point image that five width of cloth quality have nothing in common with each other, be defined as the right visual point image of A quality point, the right visual point image of B quality point, the right visual point image of C quality point, the right visual point image of D quality point, the right visual point image of E quality point respectively, be designated as R respectively
A, R
B, R
C, R
D, R
E, wherein, with the right visual point image R of A quality point
AThe basic coding quantization parameter that adopts during coding is designated as QP
RA, with the right visual point image R of B quality point
BThe basic coding quantization parameter that adopts during coding is designated as QP
RB, with the right visual point image R of C quality point
CThe basic coding quantization parameter that adopts during coding is designated as QP
RC, with the right visual point image R of D quality point
DThe basic coding quantization parameter that adopts during coding is designated as QP
RD, with the right visual point image R of E quality point
EThe basic coding quantization parameter that adopts during coding is designated as QP
RE
8.-3, with the left visual point image L of A quality point
ARight visual point image R with the A quality point
AThe stereo-picture that constitutes is as the 1st width of cloth test pattern, and is designated as S
AALeft visual point image L with the A quality point
ARight visual point image R with the B quality point
BThe stereo-picture that constitutes is as the 2nd width of cloth test pattern, and is designated as S
ABLeft visual point image L with the A quality point
ARight visual point image R with the C quality point
CThe stereo-picture that constitutes is as the 3rd width of cloth test pattern, and is designated as S
ACLeft visual point image L with the A quality point
ARight visual point image R with the D quality point
DThe stereo-picture that constitutes is as the 4th width of cloth test pattern, and is designated as S
ADLeft visual point image L with the A quality point
ARight visual point image R with the E quality point
EThe stereo-picture that constitutes is as the 5th width of cloth test pattern, and is designated as S
AE
8.-4, adopt and calculate S
DisThe identical method of characteristic vector X, calculate S
AAAdopt different weights proportion combined feature vectors, with S
AAAdopt j kind weights proportion combined feature vector to be designated as X
A, j, wherein, 1≤j≤m', m' represent the kind number of all weights proportion combinations.
8.-5, according to the support vector regression training pattern, the prediction S
AAAdopt the evaluating objective quality predicted value of all different weights proportion combined feature vectors, for S
AAAdopt j kind weights proportion combined feature vector X
A, j, with X
A, jAs the input vector of j kind weights than the support vector regression training pattern of recombination, prediction obtains X
A, jThe evaluating objective quality predicted value, be designated as Q
j, Q
j=f
j(X
A, j),
Wherein, w
OptBe the weight vector of optimum, (w
Opt)
TBe w
OptTransposed matrix, b
OptBe the bias term of optimum,
Expression X
A, jLinear function.
8.-6, with S
AAAdopt the mean value of evaluating objective quality predicted value of all different weights proportion combined feature vectors as S
AAThe objective evaluation value, be designated as Score
A,
8.-7,8.-4 employing and step to 8.-6 identical operations, obtain S
AB, S
AC, S
ADAnd S
AEThe objective evaluation value, be designated as Score respectively
B, Score
C, Score
DAnd Score
E
8.-8, calculate S
AA, S
AB, S
AC, S
ADAnd S
AEObjective score value, be designated as D respectively
1, D
2, D
3, D
4And D
5, D
1=0,
8.-9, from { D
J'| 1≤j'≤5} is D
1, D
2, D
3, D
4And D
5In the maximum objective score value of the value of finding out, be designated as D
Max, with D
MaxThe quality point of corresponding right visual point image is designated as q, the stereo-picture that is constituted by the left visual point image of the right visual point image of q quality point and A quality point then, and the picture quality difference that comes with the English alphabet order between the stereo-picture of the corresponding right visual point image of each quality point and the left visual point image formation of A quality point before the q quality point is distinguishable, the difference of the Y-PSNR of the Y-PSNR of the right visual point image of definition q quality point and the right visual point image of A quality point is as adopting QP
LHuman eye was to the discernable change step of appreciable minimum of asymmetric stereo image coding when left visual point image was encoded.
Below just utilize the inventive method that the performance that " Altmoabit " and " Doorflowers " stereo-picture carries out evaluating objective quality is compared.
Table 1 and table 2 have provided left visual point image (the i.e. left visual point image L of the A quality point of four width of cloth different qualities of four different qualities of " Altmoabit " and " Doorflowers " test pattern respectively
A) and their Y-PSNR (PSNR) and corresponding codes quantization parameter QP of the right visual point image of corresponding A, B, C, D, five quality point of E separately
RA, QP
RB, QP
RC, QP
RD, QP
RETable 3 and table 4 have provided the objective evaluation value of the stereo-picture that the right visual point image by the left visual point image of A quality point and five different quality points of " Altmoabit " and " Doorflowers " test pattern constitutes respectively.The quality of the more high explanation stereo-picture of objective evaluation value is more poor.From table 3 and table 4 as can be seen, decline along with the quality of left viewpoint or right visual point image, the objective evaluation value of stereo-picture can be more good more high, but because the objective evaluation value of different quality point is the variation tendency that low-key increases progressively, be difficult to from this variation tendency, find the critical quality point of right visual point image, therefore need further to calculate the objective score value of the right visual point image of each quality point.
Table 5 and table 6 have provided the objective score value of the stereo-picture that the right visual point image by the left visual point image of A quality point and five different quality points of " Altmoabit " and " Doorflowers " test pattern constitutes, maximizing from all objective score values easily respectively.For example, " Altmoabit " test pattern is at QP
L=23 o'clock, because the objective score value of the stereo-picture that the objective score value of the stereo-picture that the right visual point image of E quality point constitutes obviously constitutes greater than the right visual point image of other quality point, when the quality of left visual point image remains unchanged, and the quality of right visual point image is when dropping to the E quality point from the A quality point, the variation of the stereo-picture subjective perception that the mass change of the discernable right visual point image of human eye causes, because the right visual point image PSNR=41.794 of A quality point, the right visual point image PSNR=36.763 of E quality point, therefore, Ci Shi human eye is approximately 5.031dB to the discernable change step of appreciable minimum of asymmetric stereo scopic video coding.
Figure 14 has provided the schematic diagram that concerns of the quality of left visual point image of " Altmoabit " and " Doorflowers " test pattern that obtains through the inventive method and minimum discernable change step, decline (the PSNR value reduces) along with the quality of the left visual point image of stereo-picture, minimum discernable change step also can descend, and namely the difference of the PSNR of left visual point image and right visual point image diminishes.The quality of the left visual point image that obtains by subjective experiment in the patent 201010184200.9 of the quality of the left visual point image that obtains by the inventive method and the relation of minimum discernable change step and disclosed application on the 22nd September in 2010 and the relation of minimum discernable change step are very identical, illustrate that the inventive method is effective and feasible.
Y-PSNR and the coded quantization parameter of the left visual point image of the A quality point of table 1 " Altmoabit " test pattern and the right visual point image correspondence of five different quality points
Y-PSNR and the coded quantization parameter of the left visual point image of the A quality point of table 2 " Doorflowers " test pattern and the right visual point image correspondence of five different quality points
The objective evaluation value of the stereo-picture that the right visual point image by the left visual point image of A quality point and five different quality points of table 3 " Altmoabit " test pattern constitutes
The objective evaluation value of the stereo-picture that the right visual point image by the left visual point image of A quality point and five different quality points of table 4 " Doorflowers " test pattern constitutes
The objective score value of the stereo-picture that the right visual point image by the left visual point image of A quality point and five different quality points of table 5 " Altmoabit " test pattern constitutes
The objective score value of the stereo-picture that the right visual point image by the left visual point image of A quality point and five different quality points of table 6 " Doorflowers " constitutes