The content of the invention
The technical problem to be solved is to provide a kind of stereo-picture human eye based on Texture complication and just can examine
Feel distortion analysis method, its discrepancy adjustment that can be directed to different local grain complexities in stereo-picture perceives redundancy, can
Farthest compression coding efficiency is improved with compressed sensing redundancy using three-dimensional masking effect such that it is able to preferably meet non-
The needs of symmetrical stereo scopic video coding.
The present invention solve the technical scheme that adopted of above-mentioned technical problem for:A kind of stereo-picture based on Texture complication
Human eye just perceives distortion analysis method, it is characterised in that comprise the following steps:
1. using three-dimensional modeling and Software for producing, the different stereo-picture of the texture density of N single object is obtained,
And assume the single object in N width stereo-pictures texture density be from dredging to close, wherein, N>1;Then in three-dimensional modeling and system
In making software, the alpha passages of the single object in editing the left view dot image and right visual point image of every width stereo-picture each,
The single object for needing to retain is set to non-transmission region and fills pure white, remainder is set to into transmission region and is filled
Black, then render the respective mask bianry image of left view dot image and right visual point image for producing every width stereo-picture;Then
According to the respective mask bianry image of left view dot image and right visual point image of every width stereo-picture, by calculating average local side
Poor size, will be to the n-th width stereo-picture using average local reappraising the average texture complexity of every width stereo-picture
Variance calculates the average texture complexity tried to achieve and is designated as ALVn, wherein, 1≤n≤N;
2. K encoded stereoscopic image contrast test collection corresponding to every width stereo-picture is obtained, N width stereo-picture has N × K
Individual encoded stereoscopic image contrast test collection, wherein, K encoded stereoscopic image contrast test collection corresponding to the n-th width stereo-picture
Acquisition process is:
2. it is [QP _ 1, to set a left view point coded quantization parameter value intervalmin,QPmax);Then from [QPmin,
QPmax) in unique step take the different left view point coded quantization parameters of K, respectively QPL,1,QPL,2,…,QPL,K, and QPL,1=
QPmin;Wherein, QPminRepresent the minimum code quantization parameter of setting, QPmaxRepresent the maximum coded quantization parameter of setting, K>1,
QPL,1,QPL,2,…,QPL,KRepresent taken the 1st, the 2nd ..., k-th left view point coded quantization parameter, QPL,1<QPL,2
<…<QPL,K;
2. _ 2, by k-th left view point coded quantization parameter QPL,kCurrent left view point coded quantization parameter is defined as, wherein, 1
The initial value of≤k≤K, k is 1;
2. _ 3, using Video coding software, and the left side using current left view point coded quantization parameter to the n-th width stereo-picture
Visual point image is encoded with intraframe coding method, and the coding left image for obtaining is designated as into Ln,k;
2. a right viewpoint coded quantization parameter value _ 4, is set interval for [QPL,k,QPmax];Then from [QPL,k,
QPmax] in unique step take M right viewpoint coded quantization parameters, respectively QPR,1,QPR,2,…,QPR,M, and QPR,1=QPL,k;Its
In, M>1, QPR,1,QPR,2,…,QPR,MRepresent taken the 1st, the 2nd ..., the right viewpoint coded quantization parameter of m-th, QPR,1
<QPR,2<…<QPR,M;
2. _ 5, using Video coding software, and using M right viewpoint coded quantization parameter respectively to the n-th width stereo-picture
Right visual point image encoded with intraframe coding method, the coding right image of M width different qualities is obtained, will be using m-th
Right viewpoint coded quantization parameter QPR,mThe right visual point image of the n-th width stereo-picture is carried out encoding what is obtained with intraframe coding method
N-th width coding right image is designated as Rn,k,m, wherein, 1≤m≤M;
2. _ 6, by Ln,kRespectively the encoded stereoscopic image that right image constitutes M width different qualities is encoded with M width, by Ln,kWith
Rn,k,mThe encoded stereoscopic image of composition is designated as Sn,k,m;Then by Sn,k,1As reference encoder stereo-picture, by Sn,k,2,
Sn,k,3,…,Sn,k,MRespectively as Test code stereo-picture, wherein, Sn,k,1Represent Ln,kRight image R is encoded with the 1st widthn,k,1Structure
Into encoded stereoscopic image, Sn,k,2Represent Ln,kRight image R is encoded with the 2nd widthn,k,2The encoded stereoscopic image of composition, Sn,k,3Represent
Ln,kRight image R is encoded with the 3rd widthn,k,3The encoded stereoscopic image of composition, Sn,k,MRepresent Ln,kRight image R is encoded with M widthn,k,M
The encoded stereoscopic image of composition;Then by Sn,k,1Respectively with Sn,k,2,Sn,k,3,…,Sn,k,MEncoded stereoscopic image is combined into one by one
Right, the M-1 encoded stereoscopic image that combination is obtained is defined as k-th encoded stereoscopic image contrast test to the set for constituting
Collection, is designated as { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,M, wherein, Sn,k,1Sn,k,2Represent Sn,k,1With Sn,k,2It is combined into
Encoded stereoscopic image pair, Sn,k,1Sn,k,3Represent Sn,k,1With Sn,k,3The encoded stereoscopic image pair being combined into, Sn,k,1Sn,k,MRepresent
Sn,k,1With Sn,k,MThe encoded stereoscopic image pair being combined into;
2. k=k+1 _ 7, is made, then by k-th left view point coded quantization parameter QPL,kAs current left view point coded quantization
Parameter, returns again to step and 2. _ 2 continues executing with, until K left view point coded quantization parameter selects to finish, the n-th width is obtained three-dimensional
The corresponding K encoded stereoscopic image contrast test collection of image, wherein, "=" in k=k+1 is assignment;
3. organize the participant of stereo-picture subjective quality assessment several;Then adopt what is contrasted two-by-two by every participant
Mode, subjective observation and each encoded stereoscopic image contrast test collection corresponding to every width stereo-picture is contrasted on three-dimensional display
In be combined into two width encoded stereoscopic images of each encoded stereoscopic image pair, then judge that quality is passed judgment in marking;Then count all
Participant's each encoded stereoscopic image contrast test central combination corresponding to every width stereo-picture is into each encoded stereoscopic image
To two width encoded stereoscopic images fraction;
Wherein, subjective observation and k-th encoded stereoscopic image corresponding to the n-th width stereo-picture is contrasted on three-dimensional display
Contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn be combined into each encoded stereoscopic image pair two width compile
Before code stereo-picture, playing sequence is first determined, detailed process is:Would be combined into the two width coding of each encoded stereoscopic image pair
The right position randomization that stereo-picture shows on three-dimensional display, and make Flag represent reference encoder stereo-picture Sn,k,1
The position mark shown on three-dimensional display, if reference encoder stereo-picture Sn,k,1The position shown on three-dimensional display is a left side
Side, then make Flag=1;If reference encoder stereo-picture Sn,k,1The position shown on three-dimensional display is the right, then make Flag
=0;Meanwhile, by playing sequence randomization of the M-1 encoded stereoscopic image to the display on three-dimensional display;
Wherein, every participant is in the k-th encoded stereoscopic image contrast test collection corresponding to the n-th width stereo-picture
{Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn be combined into two width encoded stereoscopic images of each encoded stereoscopic image pair
When judging that quality is passed judgment in marking, two width encoded stereoscopic images of each encoded stereoscopic image pair show on three-dimensional display when
Between be T1kSecond, the playing interval time of two for playing in front and back encoded stereoscopic image pair is T2kSecond, 1s<T2k<T1k, Qian Houbo
Gray-scale maps are shown in the playing interval of two encoded stereoscopic images pair put for every participant's rest, the evaluation option of setting is
Three, respectively left image quality is good, similar, right image quality is good, if in T1kEvery participant thinks to show in second
Good in the quality of the encoded stereoscopic image on the left side of three-dimensional display, then the measured evaluation option of left image matter obtains 1 point, and poor
The measured evaluation option of few evaluation option and right image matter obtains 0 point;If in T1kEvery participant thinks to show in second
Good in the quality of the encoded stereoscopic image on the right of three-dimensional display, then the measured evaluation option of right image matter obtains 1 point, and poor
The measured evaluation option of few evaluation option and left image matter obtains 0 point;If in T1kEvery participant thinks to show in second
The left side of three-dimensional display encoded stereoscopic image and be displayed in three-dimensional display the right encoded stereoscopic image quality
Almost, then much the same evaluation option obtains 1 point, and the measured evaluation option of left image matter and right image matter is measured comments
Determine option and obtain 0 point;If in T1kSecond interior every participant cannot judge the encoded stereoscopic figure on the left side for being displayed in three-dimensional display
Picture and be displayed in three-dimensional display the right encoded stereoscopic image quality, then much the same evaluation option obtains 1 point, and the left side
The good evaluation option of picture quality and the measured evaluation option of right image matter obtain 0 point;
4. calculate each coding concentrated for each encoded stereoscopic image contrast test corresponding to every width stereo-picture to stand
The distortion of body image pair finds number and distortion probability of detection, by for k-th encoded stereoscopic figure corresponding to the n-th width stereo-picture
As contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn g-th encoded stereoscopic image to Sn,k, 1Sn,k,g-1Distortion find number and distortion probability of detection correspondence be designated as numn,k,gAnd Pn,k,gIf, Flag=1, numn,k,gFor
The measured gross score of left image matter in evaluation option;If Flag=0, numn,k,gRight image quality is good in evaluate option
Gross score;Wherein, 1≤g≤M-1, NUM represent the total number of persons of participant;
Then the encoded stereoscopic image that each encoded stereoscopic image contrast test corresponding to every width stereo-picture is concentrated is calculated
The right viewpoint coded quantization parameter that the coding right image of the Test code stereo-picture of centering is adopted is adopted relative to coding left image
The tolerable of left view point coded quantization parameter increases value range, by k-th encoded stereoscopic figure corresponding to the n-th width stereo-picture
As contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn encoded stereoscopic image centering Test code stand
The left view point encoding amount that the right viewpoint coded quantization parameter that the coding right image of body image is adopted is adopted relative to coding left image
The tolerable increase value range for changing parameter is designated as Δ QPn,k,th, Δ QPn,k,th=QPn,k,th-QPL,k, wherein,Pn,k,aRepresent for k-th volume corresponding to the n-th width stereo-picture
Code stereo-picture contrast test set { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn all encoded stereoscopic images pair
In distortion probability of detection less than 0.5 and closest to 0.5 distortion probability of detection, Pn,k,bRepresent for the n-th width stereo-picture correspondence
K-th encoded stereoscopic image contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn all codings stand
In the distortion probability of detection of body image pair more than 0.5 and closest to 0.5 distortion probability of detection, QPR,aRepresent Pn,k,aCorresponding volume
The right viewpoint coded quantization parameter that the coding right image of the Test code stereo-picture of code stereo image pair is adopted, QPR,bRepresent
Pn,k,bThe right viewpoint coded quantization that the coding right image of the Test code stereo-picture of corresponding encoded stereoscopic image centering is adopted
Parameter;
5. N number of tolerable corresponding to each left view point coded quantization parameter increases value range carries out linear fit, obtains
Each corresponding linear fit equation of left view point coded quantization parameter, by QPL,kCorresponding linear fit equation is expressed as:Δ
QPk,th=Ak×ALV+Bk, wherein, Δ QPk,thRepresent QPL,kThe JND threshold value of corresponding linear fit equation, ginseng
Number AkRepresent QPL,kThe slope of corresponding linear fit equation, parameter BkRepresent QPL,kThe intercept of corresponding linear fit equation, ginseng
Number AkWith parameter BkDirectly obtain in linear fit, ALV represents the solid for treating that human eye just perceives distortion analysis to any one width
Image calculates the average texture complexity tried to achieve using average local variance;
6. the common K linear fit equation that 5. step obtains is collectively expressed as into linear equation race formula, as:ΔQPth
=A (QPL)×ALV+B(QPL), it is minimum that the formula as can perceive stereoscopic vision based on the stereo-picture human eye of Texture complication
The quantitative relationship of discernable change quantization parameter, wherein, Δ QPthRepresent that minimum discernable change quantifies parameter threshold, A (QPL)
Represent the slope of linear equation race, B (QPL) represent the intercept of linear equation race, QPLRepresent left view point coded quantization parameter, A
(QPL)、B(QPL) it is QPLFunction, A (QPL) linear equation compiled by K left view point coded quantization parameter and K left view point
The slope linear fit of each self-corresponding linear fit equation of code quantization parameter is obtained, B (QPL) linear equation by K left view
The intercept linear fit of point coded quantization parameter and each self-corresponding linear fit equation of K left view point coded quantization parameter is obtained
Arrive.
The process of the described step 1. middle different stereo-picture of texture density for obtaining N single object is:
1. _ a1, using three-dimensional modeling and Software for producing, stereomodel is set up as single object using any one object,
The stereomodel includes background and single object model;
1. _ a2, using the even grain figure carried in three-dimensional modeling and Software for producing, the surface of single object model is entered
Row surface pinup picture, by the N number of pinup picture density rating in adjustment three-dimensional modeling and the texture editor in Software for producing, renders system
The different stereo-picture of the texture density of N single object is made, the texture on single object in every width stereo-picture
Uniformity.
Described step 1. middle ALVnAcquisition process be:
1. _ b1, using the n-th width stereo-picture left view dot image a left side of the mask bianry image to the n-th width stereo-picture
Visual point image carries out mask process, obtains left image after the n-th width mask;And using the right visual point image of the n-th width stereo-picture
Mask bianry image carries out mask process to the right visual point image of the n-th width stereo-picture, obtains right image after the n-th width mask;
1. _ b2, using the computing formula of average local variance, calculate the average local variance of left image after the n-th width mask
Value, is designated as ALVn,L,And using the computing formula of average local variance, calculate the n-th width mask
Afterwards the average local variance value of right image, is designated as ALVn,R,Wherein, PnumRepresent that the n-th width is covered
The total number of the pixel after film in left image, that is to say the total number of the pixel after the n-th width mask in right image, MBnumTable
Show that the size after the n-th width mask in left image is the total number of the image block of 8 × 8 non-overlapping copies, that is to say that the n-th width is covered
Size after film in right image for 8 × 8 non-overlapping copies image block total number, 1≤i≤MBnum, vn,L,i,8×8Represent
The variance of i-th image block after the n-th width mask in left image, vn,R,i,8×8Show i-th figure in right image after the n-th width mask
As the variance of block;
1. _ b3, the average local variance value for calculating the n-th width stereo-picture, are designated as ALV'n,
Then ALV is maden=ALV'n, that is, use ALV'nTo represent the average texture complexity of the n-th width stereo-picture, wherein, ALVn=ALV'n
In "=" be assignment.
Described step 2. _ 1 in take QPmin=20, take QPmax=51.
Described step 2. _ 4 in QPR,2=QPR,1+δk, QPR,M=QPR,1+(M-1)δk, wherein, δkRepresent in current left view
The absolute value of the difference of two neighboring right viewpoint coded quantization parameter, works as QP under point coded quantization parametermax-QPL,kDuring=M, take
δk=1;Work as QPmax-QPL,k>During M, takeRound () is round function.
Compared with prior art, it is an advantage of the current invention that:
The inventive method from Texture complication angle, the matter of right visual point image when analyzing asymmetric stereo image coding
Measure the maximum quantization parameter coding scope that can decline of quality relative to left view dot image, by a large amount of subjective experiments, with
In the case that left view point picture quality immobilizes, determining human eye can perceive the critical of the right view-point image quality of stereo-picture change
Value, obtains the largest tolerable quantization parameter threshold value of right visual point image and Texture complication and coded quantization is joined by linear fit
Quantitative mathematical model relation between number so that can be reached by reducing right view-point image quality and improve compression coding effect
The purpose of rate, while and prevent observer from perceiving the decline of right view-point image quality using stereoscopic vision masking effect, from
And ensure that the total quality of stereo-picture.
Specific embodiment
The present invention is described in further detail below in conjunction with accompanying drawing embodiment.
A kind of stereo-picture human eye based on Texture complication proposed by the present invention just perceives distortion analysis method, and its is total
Body realizes block diagram as shown in figure 1, it is comprised the following steps:
1. it is excessively complicated in view of the stereo-picture of natural scene, hardly result in only texture variations and color, background are bright
The factors such as degree, parallax, contrast keep several constant stereo-pictures, are not appropriate for for quantitative study texture information for vertical
The impact of body vision threshold of perception current, hence with existing three-dimensional modeling and Software for producing, the texture for obtaining N single object is close
The different stereo-picture of degree, and assume the single object in N width stereo-pictures texture density be from dredging to close, wherein, N
>1;Then in three-dimensional modeling and Software for producing, in editing the left view dot image and right visual point image of every width stereo-picture each
Single object alpha passages, it would be desirable to the single object of reservation is set to non-transmission region and fills pure white, by its remaining part
Set up separately and be set to transmission region and fill black, then render produce every width stereo-picture left view dot image and right visual point image it is each
From mask bianry image;Because the texture percentage ratio set in three-dimensional modeling and Software for producing does not possess versatility, therefore this
Invention is put down then according to the respective mask bianry image of left view dot image and right visual point image of every width stereo-picture by calculating
The size of local variance is only examined reappraising the average texture complexity of every width stereo-picture when Texture complication is evaluated
Consider the texture of single object and do not consider background texture, will calculate what is tried to achieve using average local variance to the n-th width stereo-picture
Average texture complexity is designated as ALVn, wherein, 1≤n≤N.
In the present embodiment, three-dimensional modeling adopts Maya2015 with Software for producing;Consider the comprehensive of subjective experiment
And workload, can use N=6.
In the present embodiment, the mistake of the step 1. middle different stereo-picture of texture density for obtaining N single object
Cheng Wei:
1. _ a1, using three-dimensional modeling and Software for producing, set up three-dimensional as single object using any one object such as bead
Model, the stereomodel includes background and single object model.
1. _ a2, using the even grain figure carried in three-dimensional modeling and Software for producing, the surface of single object model is entered
Row surface pinup picture, only changes texture close in the case where ensureing that the holding of the factors such as color, background luminance, parallax, contrast is constant
Degree, i.e., by the N number of pinup picture density rating in adjustment three-dimensional modeling and the texture editor in Software for producing, (as human eye is felt
The density grade for arriving), render the different stereo-picture of texture density for producing N single object, every width stereo-picture
In single object on texture uniformity.
Fig. 2 a left view dot images and right visual point image for giving the 1st width stereo-picture corresponding with Fig. 2 b, Fig. 3 a and Fig. 3 b
Correspondence gives the left view dot image and right visual point image of the 2nd width stereo-picture, and Fig. 4 a are corresponding with Fig. 4 b to give the 3rd width solid
The left view dot image and right visual point image of image, Fig. 5 a left view dot images for giving the 4th width stereo-picture corresponding with Fig. 5 b and
Right visual point image, Fig. 6 a left view dot images and right visual point image for giving the 5th width stereo-picture corresponding with Fig. 6 b, Fig. 7 a and figure
7b correspondences give the left view dot image and right visual point image of the 6th width stereo-picture.As can be seen that the 1st from Fig. 2 a to Fig. 7 b
Single object in width to the 6th width stereo-picture is the texture density of bead successively from dredging to close increase.
Fig. 8 a give the mask bianry image of Fig. 2 a, and Fig. 8 b give the mask bianry image of Fig. 2 b.
In the present embodiment, step 1. middle ALVnAcquisition process be:
1. _ b1, using the n-th width stereo-picture left view dot image a left side of the mask bianry image to the n-th width stereo-picture
Visual point image carries out mask process, removes background, left image after the n-th width mask is obtained, such as the left view of the 1st width stereo-picture
Dot image, mask process is carried out using Fig. 8 a to Fig. 2 a;And the mask binary map of the right visual point image using the n-th width stereo-picture
As the right visual point image to the n-th width stereo-picture carries out mask process, remove background, obtain right image after the n-th width mask, it is such as right
In the right visual point image of the 1st width stereo-picture, mask process is carried out to Fig. 2 b using Fig. 8 b.
1. _ b2, using the computing formula of average local variance (Average of Local Variance, ALV), calculate
The average local variance value of left image, is designated as ALV after n-th width maskn,L,And using average
The computing formula of local variance, calculates the average local variance value of right image after the n-th width mask, is designated as ALVn,R,Wherein, PnumThe total number of the pixel after the n-th width mask in left image is represented, be that is to say
The total number of the pixel after the n-th width mask in right image, MBnumRepresent that the size after the n-th width mask in left image is 8
The total number of the image block of × 8 non-overlapping copies, that is to say size after the n-th width mask in right image for 8 × 8 mutually not
The total number of the image block of overlap, i.e., be first respectively divided into right image after left image after the n-th width mask and the n-th width mask multiple
Size for 8 × 8 non-overlapping copies image block, 1≤i≤MBnum, vn,L,i,8×8Represent after the n-th width mask in left image
The variance of i-th image block, vn,R,i,8×8Show the variance of i-th image block after the n-th width mask in right image.
1. _ b3, the average local variance value for calculating the n-th width stereo-picture, are designated as ALV'n,
Then ALV is maden=ALV'n, that is, use ALV'nTo represent the average texture complexity of the n-th width stereo-picture, wherein, ALVn=ALV'n
In "=" be assignment.
In the present embodiment, table 1 gives the 1st width to the 6th width stereo-picture respective average texture complexity.
The width of table 1 the 1st is to the respective average texture complexity of the 6th width stereo-picture
Stereo-picture |
Average texture complexity |
1st width |
0.5503 |
2nd width |
3.3571 |
3rd width |
4.5538 |
4th width |
6.2705 |
5th width |
8.1500 |
6th width |
8.2547 |
2. K encoded stereoscopic image contrast test collection corresponding to every width stereo-picture is obtained, N width stereo-picture has N × K
Individual encoded stereoscopic image contrast test collection, wherein, K encoded stereoscopic image contrast test collection corresponding to the n-th width stereo-picture
Acquisition process is:
2. it is [QP _ 1, to set a left view point coded quantization parameter value intervalmin,QPmax);Then from [QPmin,
QPmax) in unique step take the different left view point coded quantization parameters of K, respectively QPL,1,QPL,2,…,QPL,K, and QPL,1=
QPmin;Wherein, QPminRepresent the minimum code quantization parameter of setting, QPmaxRepresent the maximum coded quantization parameter of setting, K>1,
QPL,1,QPL,2,…,QPL,KRepresent taken the 1st, the 2nd ..., k-th left view point coded quantization parameter, QPL,1<QPL,2
<…<QPL,K。
In the present embodiment, QP is takenmin=20, take QPmax=51;Consider left view point coded quantization ginseng in subjective experiment
The comprehensive and workload of number value, takes K=4, such QPL,1,QPL,2,QPL,3,QPL,4It is 20,26,32,38 that value can be corresponded to,
For the left view dot image of N width stereo-pictures, K different left view point coded quantization parameter of selection all, i.e., for N width
The left view dot image of stereo-picture, the 1st left view point coded quantization parameter of selection is identical, the like.
2. _ 2, by k-th left view point coded quantization parameter QPL,kCurrent left view point coded quantization parameter is defined as, wherein, 1
The initial value of≤k≤K, k is 1.
2. _ 3, using existing Video coding software, and using current left view point coded quantization parameter to the n-th width axonometric chart
The left view dot image of picture is encoded with intraframe coding method, and the coding left image for obtaining is designated as into Ln,k。
In the present embodiment, Video coding software can adopt HEVC video encoding standard reference softwares HM14.0.
2. a right viewpoint coded quantization parameter value _ 4, is set interval for [QPL,k,QPmax];Then from [QPL,k,
QPmax] in unique step take M right viewpoint coded quantization parameters, respectively QPR,1,QPR,2,…,QPR,M, and QPR,1=QPL,k;Its
In, M>1, QPR,1,QPR,2,…,QPR,MRepresent taken the 1st, the 2nd ..., the right viewpoint coded quantization parameter of m-th, QPR,1
<QPR,2<…<QPR,M。
In the present embodiment, the comprehensive and work of right viewpoint coded quantization parameter value in subjective experiment is considered
Amount, takes M=13.
In the present embodiment, step 2. _ 4 in QPR,2=QPR,1+δk, QPR,M=QPR,1+(M-1)δk, wherein, δkRepresent
The absolute value of the difference of two neighboring right viewpoint coded quantization parameter, works as QP under current left view point coded quantization parametermax-QPL,k
During=M, δ is takenk=1;Work as QPmax-QPL,k>During M, takeRound () is round
Function.
Table 2 gives QPL,1,QPL,2,QPL,3,QPL,4Under each self-corresponding M right viewpoint coded quantization parameters.
Table 2QPL,1,QPL,2,QPL,3,QPL,4Under each self-corresponding M right viewpoint coded quantization parameters
2. it is _ 5, using existing Video coding software and vertical to the n-th width respectively using M right viewpoint coded quantization parameter
The right visual point image of body image is encoded with intraframe coding method, and the coding right image of M width different qualities is obtained, and will be adopted
M-th right viewpoint coded quantization parameter QPR,mThe right visual point image of the n-th width stereo-picture is encoded with intraframe coding method
The the n-th width coding right image for obtaining is designated as Rn,k,m, wherein, 1≤m≤M.
2. _ 6, by Ln,kRespectively the encoded stereoscopic image that right image constitutes M width different qualities is encoded with M width, by Ln,kWith
Rn,k,mThe encoded stereoscopic image of composition is designated as Sn,k,m;Due to stereo-picture left view point picture coding when the left view point that adopts compile
When the right viewpoint coded quantization parameter that adopts is equal when the right visual point image of code quantization parameter and stereo-picture is encoded, the volume for obtaining
Code stereo-picture it is best in quality, therefore and then by Sn,k,1As reference encoder stereo-picture, by Sn,k,2,Sn,k,3,…,Sn,k,M
Respectively as Test code stereo-picture, wherein, Sn,k,1Represent Ln,kRight image R is encoded with the 1st widthn,k,1The encoded stereoscopic of composition
Image, Sn,k,2Represent Ln,kRight image R is encoded with the 2nd widthn,k,2The encoded stereoscopic image of composition, Sn,k,3Represent Ln,kCompile with the 3rd width
Code right image Rn,k,3The encoded stereoscopic image of composition, Sn,k,MRepresent Ln,kRight image R is encoded with M widthn,k,MThe coding of composition stands
Body image;Then by Sn,k,1Respectively with Sn,k,2,Sn,k,3,…,Sn,k,MEncoded stereoscopic image pair is combined into one by one, and combination is obtained
M-1 encoded stereoscopic image to constitute set be defined as k-th encoded stereoscopic image contrast test collection, be designated as { Sn,k, 1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,M, wherein, Sn,k,1Sn,k,2Represent Sn,k,1With Sn,k,2The encoded stereoscopic figure being combined into
As right, Sn,k,1Sn,k,3Represent Sn,k,1With Sn,k,3The encoded stereoscopic image pair being combined into, Sn,k,1Sn,k,MRepresent Sn,k,1With Sn,k,MGroup
The encoded stereoscopic image pair of synthesis.
2. k=k+1 _ 7, is made, then by k-th left view point coded quantization parameter QPL,kAs current left view point coded quantization
Parameter, returns again to step and 2. _ 2 continues executing with, until K left view point coded quantization parameter selects to finish, the n-th width is obtained three-dimensional
The corresponding K encoded stereoscopic image contrast test collection of image, wherein, "=" in k=k+1 is assignment.
3. organize the participant of stereo-picture subjective quality assessment several;Then adopt what is contrasted two-by-two by every participant
Mode, subjective observation and each encoded stereoscopic image contrast test collection corresponding to every width stereo-picture is contrasted on three-dimensional display
In be combined into two width encoded stereoscopic images of each encoded stereoscopic image pair, then judge that quality is passed judgment in marking;Then count all
Participant's each encoded stereoscopic image contrast test central combination corresponding to every width stereo-picture is into each encoded stereoscopic image
To two width encoded stereoscopic images fraction.
In the present embodiment, the participant of experiment has 20,12 male 8 female, and the mean age was 25 one full year of life, and everyone vision is just
Often or correct defects of vision normal.
In the present embodiment, subjective observation is carried out on three-dimensional display, and the stereoscopic display device for such as using is three
Star UA65F9000AJ ultra high-definition UHD shutter stereoscopic televisions (65 inches, 16:9), the stereo display spatial resolution of display
For 1920x1080;During viewing, participant needs to wear shutter stereo spectacles, and the distance watched is about 4 times of image
Highly, about 2.8 meters.
K-th encoded stereoscopic image contrast test collection { S corresponding to the n-th width stereo-picture is shown on three-dimensional displayn,k, 1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn when being combined into two width encoded stereoscopic image of each encoded stereoscopic image pair, if
Sn,k,1All the time the left side of three-dimensional display, and S are displayed inn,k,2、Sn,k,3、…、Sn,k,MQuality changed successively to differing from by good, then
Vision inertia, and a certain degree of psychological hint occurs unavoidably.Therefore the present invention is in order to avoid subjective observation contrast experiment
In vision inertia, more objectively reflect experimental result, need before subjective comparative's experiment first to { Sn,k,1Sn,k,2,Sn,k, 1Sn,k,3,…,Sn,k,1Sn,k,MIn be combined into two width encoded stereoscopic images of each encoded stereoscopic image pair on three-dimensional display
Display location and { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn be combined into the display of each encoded stereoscopic image pair
Order is rearranged.The subjective observation and to contrast corresponding k-th coding of the n-th width stereo-picture vertical i.e. on three-dimensional display
Body image comparison test set { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn be combined into each encoded stereoscopic image pair
Before two width encoded stereoscopic images, playing sequence is first determined, detailed process is:Would be combined into the two of each encoded stereoscopic image pair
The right position randomization that width encoded stereoscopic image shows on three-dimensional display, i.e. reference encoder stereo-picture Sn,k,1Go out at random
Now the left side or the right of three-dimensional display, for later stage statistics conveniently, and makes Flag represent reference encoder stereo-picture Sn,k,1
The position mark shown on three-dimensional display, if reference encoder stereo-picture Sn,k,1The position shown on three-dimensional display is
The left side, then make Flag=1;If reference encoder stereo-picture Sn,k,1The position shown on three-dimensional display is the right, then make
Flag=0;Meanwhile, by playing sequence randomization of the M-1 encoded stereoscopic image to the display on three-dimensional display.
Table 3 corresponding to the 1st width stereo-picture gives at the 1st encoded stereoscopic image contrast test collection (i.e. QPL,1When=20)
In M-1 encoded stereoscopic image on three-dimensional display show playing sequence.
The M-1 encoded stereoscopic figure that 1st encoded stereoscopic image contrast test corresponding to the width stereo-picture of table 3 the 1st is concentrated
As to the playing sequence shown on the three-dimensional display
Every participant is in the k-th encoded stereoscopic image contrast test collection { S corresponding to the n-th width stereo-picturen,k, 1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn be combined into each encoded stereoscopic image pair two width encoded stereoscopic images judge
When quality is passed judgment in marking, the time that two width encoded stereoscopic images of each encoded stereoscopic image pair show on three-dimensional display is
T1kSecond, the playing interval time of two for playing in front and back encoded stereoscopic image pair is T2kSecond, 1s<T2k<T1k, in the present embodiment
In take T1k=10s, T2k=3s, shows that gray-scale maps supply per in the playing interval of two for playing in front and back encoded stereoscopic image pair
Participant rests, and the evaluation option of setting is three, and respectively left image quality is good, similar, right image quality is good, if
In T1kSecond interior every participant thinks that the quality of the encoded stereoscopic image on the left side for being displayed in three-dimensional display is good, then left side figure
As the measured evaluation option of matter obtains 1 point, and much the same evaluation option and the measured evaluation option of right image matter obtain 0 point;
If in T1kSecond interior every participant thinks that the quality of the encoded stereoscopic image on the right for being displayed in three-dimensional display is good, then the right
The good evaluation option of picture quality obtains 1 point, and much the same evaluation option and the measured evaluation option of left image matter obtain 0
Point;If in T1kEvery participant thinks the encoded stereoscopic image on the left side for being displayed in three-dimensional display and is displayed in solid in second
Almost, then much the same evaluation option obtains 1 point for the quality of the encoded stereoscopic image on the right of display, and left image quality
The measured evaluation option of good evaluation option and right image matter obtains 0 point;If in T1kEvery participant cannot judge to show in second
Show the matter of the encoded stereoscopic image on the left side of three-dimensional display and the encoded stereoscopic image on the right for being displayed in three-dimensional display
Amount, then be to ensure accuracy, makes much the same evaluation option obtain 1 point, and the measured evaluation option of left image matter and the right figure
As the measured evaluation option of matter obtains 0 point.
Table 4 gives the 1st encoded stereoscopic image contrast test central combination corresponding to the 1st width stereo-picture and compiles into each
The marking table of two width encoded stereoscopic images of code stereo pairs.
1st encoded stereoscopic image contrast test central combination corresponding to the width stereo-picture of table 4 the 1st is into each encoded stereoscopic
The marking table of two width encoded stereoscopic images of image pair
Table 5 gives 20 participant's 1st encoded stereoscopic image contrast tests corresponding to the 1st width stereo-picture and concentrates
It is combined into the statistical result of the two width encoded stereoscopic images marking of each encoded stereoscopic image pair.
5 20 participants of table, the 1st encoded stereoscopic image contrast test central combination corresponding to the 1st width stereo-picture
Into the statistical result that two width encoded stereoscopic images of each encoded stereoscopic image pair are given a mark
4. calculate each coding concentrated for each encoded stereoscopic image contrast test corresponding to every width stereo-picture to stand
The distortion of body image pair finds number and distortion probability of detection, by for k-th encoded stereoscopic figure corresponding to the n-th width stereo-picture
As contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn g-th encoded stereoscopic image to Sn,k, 1Sn,k,g-1Distortion find number and distortion probability of detection correspondence be designated as numn,k,gAnd Pn,k,gIf, Flag=1, numn,k,gFor
The measured gross score of left image matter in evaluation option;If Flag=0, numn,k,gRight image quality is good in evaluate option
Gross score;Because " similar " illustrates left image and right image apparently indifference, distortion can not be found, therefore not arranged
Enter to calculate;Wherein, 1≤g≤M-1, NUM represent the total number of persons of participant.
QP is pressed to table 5R,mOrder from small to large is rearranged, and then calculated distortion finds number and distortion probability of detection,
As a result as listed by table 6.
The distortion of each encoded stereoscopic image pair in the table 5 of table 6 finds number and distortion probability of detection
Then the encoded stereoscopic image that each encoded stereoscopic image contrast test corresponding to every width stereo-picture is concentrated is calculated
The right viewpoint coded quantization parameter that the coding right image of the Test code stereo-picture of centering is adopted is adopted relative to coding left image
The tolerable of left view point coded quantization parameter increases value range, by k-th encoded stereoscopic figure corresponding to the n-th width stereo-picture
As contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn encoded stereoscopic image centering Test code stand
The left view point encoding amount that the right viewpoint coded quantization parameter that the coding right image of body image is adopted is adopted relative to coding left image
The tolerable increase value range for changing parameter is designated as Δ QPn,k,th, Δ QPn,k,th=QPn,k,th-QPL,k, wherein,Pn,k,aRepresent for k-th volume corresponding to the n-th width stereo-picture
Code stereo-picture contrast test set { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn all encoded stereoscopic images pair
In distortion probability of detection less than 0.5 and closest to 0.5 distortion probability of detection, Pn,k,bRepresent for the n-th width stereo-picture correspondence
K-th encoded stereoscopic image contrast test collection { Sn,k,1Sn,k,2,Sn,k,1Sn,k,3,…,Sn,k,1Sn,k,MIn all codings stand
In the distortion probability of detection of body image pair more than 0.5 and closest to 0.5 distortion probability of detection, QPR,aRepresent Pn,k,aCorresponding volume
The right viewpoint coded quantization parameter that the coding right image of the Test code stereo-picture of code stereo image pair is adopted, QPR,bRepresent
Pn,k,bThe right viewpoint coded quantization that the coding right image of the Test code stereo-picture of corresponding encoded stereoscopic image centering is adopted
Parameter.
Such as in table 6, Pn,k,a=0.2, Pn,k,b=0.65, QPR,a=30, QPR,b=32, QPn,k,th=31.3, Δ QPn,k,th
=11.3.
It is considered herein that have half participant to think to be found that distortion in subjective marking experiment, and second half participant thinks
The situation for not finding distortion is critical viewpoint, then distortion probability of detection is 50% when reaching critical viewpoint, due to 50%
The corresponding right viewpoint coded quantization parameter of distortion probability of detection be not necessarily experiment used in coded quantization parameter, therefore this
Bright utilization linear interpolation calculates Δ QPn,k,th。
Table 7 gives what each self-corresponding 4 encoded stereoscopic image contrast tests of 6 width stereo-pictures listed by table 1 were concentrated
The right viewpoint coded quantization parameter that the coding right image of the Test code stereo-picture of encoded stereoscopic image centering is adopted relative to
The tolerable of the left view point coded quantization parameter that coding left image is adopted increases value range.
The right viewpoint coded quantization parameter that the coding right image of the Test code stereo-picture of table 7 is adopted is relative to coding left figure
As the tolerable of the left view point coded quantization parameter for adopting increases value range
5. N number of tolerable corresponding to each left view point coded quantization parameter increases value range carries out linear fit, obtains
Each corresponding linear fit equation of left view point coded quantization parameter, by QPL,kCorresponding linear fit equation is expressed as:Δ
QPk,th=Ak×ALV+Bk, wherein, Δ QPk,thRepresent QPL,kThe JND threshold value of corresponding linear fit equation, ginseng
Number AkRepresent QPL,kThe slope of corresponding linear fit equation, parameter BkRepresent QPL,kThe intercept of corresponding linear fit equation, ginseng
Number AkWith parameter BkDirectly obtain in linear fit, ALV represents the solid for treating that human eye just perceives distortion analysis to any one width
Image calculates the average texture complexity tried to achieve using average local variance.
In the present embodiment, Ak、BkThe preparation method of parameter is:For table 7, work as QPL,1When=20, N width stereo-pictures
Texture complication with to QPL,1N number of tolerable in=20 respective columns increases value range and constitutes N group data:(0.5503,11.3),
(3.3571,14.8), (4.5538,16.7), (6.2705,18.6), (8.1500,20.1), (8.2547,20.1), to the N groups
Data carry out linear fit using classical linear fit mathematical method, you can obtain QPL,1Linear fit equation when=20 and
Parameter A1、B1Value;Work as QPL,2When=26, the Texture complication of N width stereo-pictures with to QPL,2N number of in=26 respective columns can
Tolerance increases value range and constitutes N group data:(0.5503,7.6), (3.3571,10.4), (4.5538,11.2), (6.2705,
12.8), (8.1500,14.1), (8.2547,14.3), line is entered using classical linear fit mathematical method to the N groups data
Property fitting, you can obtain QPL,2Linear equation and parameter A when=262、B2Value;The like, 4 linear fits are obtained
Equation, the slope A of each linear fit equationkWith intercept BkAs listed by table 8,4 corresponding 4 of linear fit equations are fitted song
Line is as shown in Figure 9.
The slope and intercept of table 8 each linear fit equation
6. the common K linear fit equation that 5. step obtains is collectively expressed as into linear equation race formula, as:ΔQPth
=A (QPL)×ALV+B(QPL), it is minimum that the formula as can perceive stereoscopic vision based on the stereo-picture human eye of Texture complication
The quantitative relationship of discernable change quantization parameter, wherein, Δ QPthRepresent that minimum discernable change quantifies parameter threshold, A (QPL)
Represent the slope of linear equation race, B (QPL) represent the intercept of linear equation race, QPLRepresent left view point coded quantization parameter, A
(QPL)、B(QPL) it is QPLFunction, A (QPL) linear equation compiled by K left view point coded quantization parameter and K left view point
The slope linear fit of each self-corresponding linear fit equation of code quantization parameter is obtained, B (QPL) linear equation by K left view
The intercept linear fit of point coded quantization parameter and each self-corresponding linear fit equation of K left view point coded quantization parameter is obtained
Arrive.
4 groups of data are constituted using 4 left view point coded quantization parameters in table 8 and corresponding 4 slopes:(20,
1.1495) (26,0.8441) (32, and 0.5905) (38,0.0614), to 4 groups of data using classical linear fit mathematics side
Method carries out linear fit, obtains linear equation A (QPL)=p1QPL+q1, wherein, parameter p1Represent the slope of equation, parameter q1Table
Show the intercept of equation, parameter p can be obtained by linear fit1、q1Respectively -0.0586,2.3601.
4 groups of data are constituted using 4 left view point coded quantization parameters in table 8 and corresponding 4 intercepts:(20,
10.9790) (26,7.3383) (32, and 3.3962) (38,2.7381), to 4 groups of data using classical linear fit mathematics side
Method carries out linear fit, obtains linear equation B (QPL)=p2QPL+q2, wherein, parameter p2Represent the slope of equation, parameter q2Table
Show the intercept of equation, parameter p can be obtained by linear fit2、q2Respectively -0.4782,19.9808.
For any one width treats that human eye just perceives the stereo-picture of distortion analysis, the left view point coding that its left view point is adopted
Quantization parameter is, it is known that A (QPL) and B (QPL), it is known that and its average texture complexity, it is known that therefore any one width can be calculated
Treat that human eye just perceives the discernable change of minimum of the stereo-picture of distortion analysis and quantifies parameter threshold.
Figure 10 a give A (QPL) fit linear relationship, Figure 10 b give B (QPL) fit linear relationship, Figure 11 gives
Linear equation race Δ QP is gone outth=A (QPL)×ALV+B(QPL) surface chart.