CN102982535A - Stereo image quality evaluation method based on peak signal to noise ratio (PSNR) and structural similarity (SSIM) - Google Patents
Stereo image quality evaluation method based on peak signal to noise ratio (PSNR) and structural similarity (SSIM) Download PDFInfo
- Publication number
- CN102982535A CN102982535A CN2012104311231A CN201210431123A CN102982535A CN 102982535 A CN102982535 A CN 102982535A CN 2012104311231 A CN2012104311231 A CN 2012104311231A CN 201210431123 A CN201210431123 A CN 201210431123A CN 102982535 A CN102982535 A CN 102982535A
- Authority
- CN
- China
- Prior art keywords
- psnr
- image
- stereo
- value
- ssim
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention belongs to the image processing field, provides an effective and objective quality evaluation method, and has a certain promoting effect on the development of stereo imaging technique. According to the technique scheme, a stereo image quality evaluation method based on a peak signal to noise ratio (PSNR) and structural similarity (SSIM) comprises obtaining an absolute difference image of an original image; calculating weighted average structural similarity by utilizing the absolute parallax image; extracting parallax of the stereo image by adopting stereo matching algorithm based on picture segmentation; calculating the PSNR of the difference image of the original stereo image and a difference image of a distortion stereo image; calculating the PSNR of the original stereo image X and the distortion stereo image Y; calculating an initial evaluation value of the stereo image distortion degree; calculating the PSNR of parallax information of the distortion stereo image and the parallax information of the original stereo image; and finally a quality evaluation value is obtained. The stereo image quality evaluation method based on the PSNR and the SSIM is mainly used for image processing.
Description
Technical field
The invention belongs to image processing field, relate to stereo image quality objective evaluation algorithm, relate in particular to a kind of stereo image quality evaluation method based on PSNR and SSIM.
Background technology
Stereo-picture is different from plane picture, exists the correlativity of height between the adjacent viewpoint of stereo-picture.If the picture quality of two adjacent viewpoint is all very high, but the parallax between viewpoint is less, the image stereoscopic sensation that the observer feels can reduce.Therefore, existing plane picture evaluating objective quality algorithm can not directly apply to the three-dimensional image objective quality evaluation algorithm.
Usually based on the more serious thought of the larger quality degradation of gray difference of standard picture, objective evaluation index relatively more commonly used is square error (MSE) and Y-PSNR (PSNR) to traditional image evaluating objective quality at present.Given width of cloth size is digitized image f (x, y) and the reference picture f of M * N
0(x, y), then the PSNR of image f is defined as:
Wherein, f
MaxThe maximum gradation value of function f (x, y), the gray level image f of 8bit commonly used
MaxValue be 255.
So far, also do not have unified three-dimensional image objective evaluation index, relatively more commonly used is the Y-PSNR (PPSNR) of differential chart between the viewpoint of the left and right sides, is defined as:
Wherein, f (x, y)
l, f (x, y)
lRepresent respectively the left and right sides view of Given Graph picture, f
0(x, y)
l, f
0(x, y)
rRepresent the left and right sides view of reference picture.
Theoretical (the SSIM of structural similarity, str people ctural similarity) is a kind of new method about image quality evaluation, the people such as WangZhou and Bovik for many years image is processed and basis that image quality evaluation is studied on, the concept of structural information was proposed for the first time in international conference in 2002, and in more detailed description in 2004 the structure similarity theory.The angle that SSIM forms from image is defined as the attribute that is independent of brightness, contrast, can reflects object structures the scene with the structural information of image, and structure distortion is modeled as the combination of brightness, contrast and three different factors of structure.Think that based on the image quality evaluating method of structural similarity the major function of human eye vision is to extract the structural information of image, and the human visual system can realize this goal height adaptive, so image quality evaluation can be thought the tolerance to the picture structure distortion approx.
The block diagram of Fig. 1 for utilizing structural similarity (SSIM) to carry out image quality evaluation, the original image that X is (signal), Y are distorted image (signal).
Structural similarity tolerance is the combination of three different factors with distortion modeling: brightness (l), contrast (c) and structure (s).With average (μ
X, μ
Y) as the estimation of brightness, use standard deviation (σ
X, σ
Y) the as a comparison estimation of degree, use covariance sigma
XyAs the tolerance of structural similarity degree, its computing formula is defined as:
Structural similarity tolerance is defined as:
SSIM(X,Y)=[l(X,Y)]
α[c(X,Y)]
β[s(X,Y)]
γ (2)
In formula (1) and the formula (2), α, beta, gamma>0, α, beta, gamma are used for adjusting the weight of brightness, contrast and structural information.c
1, c
2, c
3Being very little constant, is to avoid producing when denominator is zero or approaching zero the constant that wild effect is added.
If get c
3=c
2/ 2, α=β=γ=1, formula (1) can be reduced to:
Can estimate SSIM (except the marginal point) value of each point correspondence image piece by formula (2), SSIM is mapped as piece image and describes the quality information that is evaluated image, claims that this width of cloth image is the SSIM key map.At last, measure as the evaluation to overall image quality with average structure similarity (MSSIM).
Detailed process is: original reference image X, distorted image Y are carried out the piecemeal of non-overlapping copies with the window of formed objects, and piecemeal adds up to M
Sum, utilize moving window operation that window is moved to the lower right corner from the upper left corner by pixel ground along image, calculate the SSIM value of the corresponding subimage of each window, the SSIM value of all subimages is averaging:
In the plane picture quality assessment, structural similarity and calculation of correlation rule thereof are because the simple advantage such as efficient is subject to the extensive concern of Chinese scholars, and has been applied to some relevant fields after the proposition.
Summary of the invention
The present invention is intended to overcome the deficiencies in the prior art, and effective, objective quality evaluating method is provided, and certain impetus is played in the development of stereoscopic imaging technology.For achieving the above object, the technical scheme that the present invention takes is that the stereo image quality evaluation method based on PSNR and SSIM comprises the following steps:
The first step, with the viewpoint of original image X to (L1, R1) and the viewpoint of distorted image Y to (L2, R2) carry out respectively the phase reducing, obtain its absolute difference image Diff (X), Diff (Y), wherein, Diff (X)=| L1-R1|, Diff (Y)=| L2-R2|;
Second step utilizes the different image calculation weighted average construction of absolute parallax similarity WMSSIM, is designated as 3D_SSIM;
In the 3rd step, adopt the parallax that extracts stereo-picture based on the Stereo Matching Algorithm of image segmentation;
The 4th step is by formula PSNR
1=(PSNR
R+ PSNR
G+ PSNR
BThe Y-PSNR value PSNR of the error image Diff (X) of the original stereographic map of)/3 calculating and the error image Diff (Y) of distortion stereographic map
1, PSNR
R, PSNR
G, PSNR
BBe respectively the PSNR value of three primary color components;
The 5th step is by formula PSNR
2=(PPSNR
R+ PPSNR
G+ PPSNR
BThe Y-PSNR value PSNR between original stereo-picture X and the distortion stereo-picture Y is calculated in)/3
2, wherein, PPSNR
R, PPSNR
G, PPSNR
BBe respectively the PPSNR value of three primary color components;
The 6th step, the initial evaluation value S_VALUE of calculating stereo-picture distortion level, computing formula is: S_VALUE=f
Normalization(PSNR
1)+f
Normalization(PSNR
2)+f
Normalization(3D_SSIM));
The 7th step, after the parallax information of the parallax information of stereo-picture after the distortion and original stereo-picture extracted, calculating Y-PSNR DPSNR between the two;
The 8th step, judge the distortion level of stereo-picture according to the size of DPSNR value, according to large young pathbreaker's distorted image of DPSNR value according to table 1 classification, obtain the distorted image of different brackets, and then distribute to correction factor corresponding to each grade according to table 2, with correction factor initial evaluation value S_VALUE is revised, finally obtain quality assessment value 3D_VAL;
Table 1 image fault grading
Grade | 1 | 2 | 3 | 4 | 5 |
DPSNR | <10 | [10,12] | (12,15) | [15,20] | >20 |
The correction factor of table 2 different brackets
Grade | 1 | 2 | 3 | 4 | 5 |
Correction factor | 0.5 | 0.4 | 0.28 | 0.22 | 0.2 |
In the 9th step, the stereoscopic image quality is estimated: the 3D_VAL value is larger, and the quality of image is better.
The second step concrete steps are as follows:
1. the antipode image with stereo-picture carries out equal-sized division, is divided into B sub-block, and each sub-block is designated as B
i, pixel count is N in each sub-block
Sum
2. pass through formula
Calculate each sub-block B
iLuminance weights factor of influence v
i, grain details factor of influence t
i, locus factor of influence p
i, wherein, I
Imax=max{I
1, I
2..., I
NBe the maximum brightness value in i the sub-block district, I
kBe brightness value,
Be the average brightness value in i sub-block district, x
Io, y
IoThe center of expression piecemeal i, x
c, y
cThe centre coordinate of expression original image, P represents that each point is to the ultimate range of centre coordinate in the original image;
3. pass through formula
Determine the weight coefficient w of each sub-block
i, wherein, W
iThe factor of influence of each sub-block after comprehensive, W
SumThe combined influence factor W of all sub-blocks
iSummation, B is sub-block sum;
4. pass through formula
Calculate each sub-block B
iWeighted average construction similarity evaluation value;
5. distinguish R, G, the weighted average construction similarity value of B component: the WMSSIM of calculated difference image
R(X, Y), WMSSIM
G(X, Y), WMSSIM
B(X, Y) is by obtaining parameter 3D_SSIM, 3D_SSIM=E (WMSSIM after the formula statistical average
R(X, Y)+WMSSIM
G(X, Y)+WMSSIM
B(X, Y)).
In the 3rd step, concrete steps are followed successively by: color segmentation, Window match, the extraction of parallax plane, parallax plane optimizing, parallax plane generate.
Beneficial effect of the present invention is: the great many of experiments simulation result shows, the three-dimensional image objective quality evaluation method of the PSNR associating SSIM that the present invention proposes can effectively utilize the steric information feature, be fit to the evaluation of multiple distortion stereo image quality, and objective evaluation result and subjective assessment result's consistance is better.
Description of drawings
Fig. 1 is structural similarity tolerance block diagram.
Fig. 2 is the three-dimensional image objective quality evaluation method block diagram of PSNR associating SSIM.
Fig. 3 is that parallax extracts block diagram.
Fig. 4 is that 3D WINDOWS-19A01 type Computerized 3 D imaging device reaches supporting with it anaglyph spectacles.
Fig. 5 is double vision point stereo-picture teddy.
Fig. 6 is the distorted image of the left viewpoint of stereo-picture teddy.(a) radial blur, (b) profile adds blackly, and (c) edge strengthening (d) adds hacures, (e) adds Gaussian noise, (d) adds mosaic, (f) introduces perpendicular displacement.
Fig. 7 is the antipode figure of stereo-picture teddy.
Fig. 8 is the disparity map of original stereo-picture teddy.(a) disparity map that obtains of algorithm; (b) standard disparity map.
Fig. 9 is the evaluating objective quality value curve of stereo-picture in the different distortion situations.
Figure 10 is the scatter diagram of this algorithm and DMOS.
Figure 11 is the scatter diagram of PSNR and DMOS.
Embodiment
The present invention mainly launches the key issues such as impact of picture quality around human visual system's formation mechanism, treatment mechanism, stereoscopic vision characteristic.By to the researching and analysing of characteristics of image and visual information characteristic, designed and finished a large amount of experimental verifications, the method for evaluating objective quality of a kind of method for objectively evaluating that can be used for weighing polytype distorted image quality-PSNR associating SSIM has been proposed.Main innovate point is: the impact of the Factors on Human eye perception stereoeffects such as the brightness of consideration stereo-picture, grain details, locus, the most influential two factors of stereoscopic image imaging technique have been extracted: absolute difference and binocular vision difference information, in conjunction with the most effective Y-PSNR (PSNR, Peak Signal to Noise Ratio) and weighted average construction similarity method (WMSSIM:Weighted Mean structural similarity) in the present evaluation field.
The evaluation method of the PSNR associating SSIM of the design's proposition has been utilized based on the SSIM of human-eye visual characteristic theoretical, by calculating WSSIM and the PSNR value of stereo-picture characteristic information, obtains an initial image quality evaluation values; Then further extract the binocular parallax information of stereo-picture, in order to revise initial value; By revising, finally obtain an evaluating objective quality value.Its evaluation method block diagram as shown in Figure 2.The present invention adopts following technical scheme:
The first step, with the viewpoint of original image X to (L1, R1) and the viewpoint of distorted image Y to (L2, R2) carry out respectively the phase reducing, obtain its absolute difference image Diff (X), Diff (Y), wherein, Diff (X)=| L1-R1|, Diff (Y)=| L2-R2|.
Second step utilizes the different image calculation weighted average construction of absolute parallax similarity WMSSIM, is designated as 3D_SSIM, and concrete steps are as follows:
1. the antipode image with stereo-picture carries out equal-sized division, is divided into B sub-block, and each sub-block is designated as B
i, pixel count is N in each sub-block
Sum
2. pass through formula
Calculate each sub-block B
iLuminance weights factor of influence v
i, grain details factor of influence t
i, locus factor of influence p
iWherein, I
Imax=max{I
1, I
2..., I
NBe the maximum brightness value in i the sub-block district, I
kBe brightness value,
Be the average brightness value in i sub-block district, x
Io, y
IoThe center of expression piecemeal i, x
c, y
cThe centre coordinate of expression original image, P represents that each point is to the ultimate range of centre coordinate in the original image.
3. pass through formula
Determine the weight coefficient w of each sub-block
iWherein, W
iThe factor of influence of each sub-block after comprehensive, W
SumThe combined influence factor W of all sub-blocks
iSummation, B is sub-block sum.
4. pass through formula
Calculate each sub-block B
iWeighted average construction similarity evaluation value.
5. distinguish the average weighted structural similarity value WMSSIM (X, Y) of calculated difference image RGB component, obtain WMSSIM
R(X, Y), WMSSIM
G(X, Y), WMSSIM
B(X, Y) is by formula 3D_SSIM=E (WMSSIM
R(X, Y)+WMSSIM
G(X, Y)+WMSSIM
B(X, Y)) obtain parameter 3D_SSIM after the statistical average.
In the 3rd step, adopt the parallax that extracts stereo-picture based on the Stereo Matching Algorithm of image segmentation.Algorithm is realized block diagram as shown in Figure 3.
The 4th step is by formula PSNR
1=(PSNR
R+ PSNR
G+ PSNR
BThe Y-PSNR value PSNR of the error image Diff (X) of the original stereographic map of)/3 calculating and the error image Diff (Y) of distortion stereographic map
1PSNR
R, PSNR
G, PSNR
BBe respectively the PSNR value of three primary color components.
The 5th step is by formula PSNR
2=(PPSNR
R+ PPSNR
G+ PPSNR
BThe Y-PSNR value PSNR between original stereo-picture X and the distortion stereo-picture Y is calculated in)/3
2Wherein, PPSNR
R, PPSNR
G, PPSNR
BBe respectively the PPSNR value of three primary color components.
The 6th step, the initial evaluation value S_VALUE of calculating stereo-picture distortion level, computing formula is: S_VALUE=f
Normalization(PSNR
1)+f
Normalization(PSNR
2)+f
Normalization(3D_SSIM)).
The 7th step, after the parallax information of the parallax information of stereo-picture after the distortion and original stereo-picture extracted, calculating Y-PSNR DPSNR between the two.
In the 8th step, according to the distortion level of the size of DPSNR value judgement stereo-picture,, obtain the distorted image of different brackets, and then distribute to correction factor corresponding to each grade according to table 2 according to table 1 classification according to large young pathbreaker's distorted image of DPSNR value.With correction factor initial evaluation value S_VALUE is revised, finally obtained the quality assessment value 3D_VAL that calculates by this algorithm.
In the 9th step, the stereoscopic image quality is estimated: the 3D_VAL value is larger, and the quality of image is better.
Further describe the present invention below in conjunction with embodiment.
The design utilizes the data of U.S. Middlebury stereo datasets to carry out emulation experiment.In the experiment, we have selected 15 width of cloth double vision point stereo-pictures pair, and the image size is 436 * 360 * 24, comprises that teddy, baby3, Art, Dolls, tsukuba etc. are as test pattern.First the left view of every width of cloth viewpoint centering carried out the distortion processing, obtain the stereo-picture in a series of different distortion situations.For the performance of verification system more intuitively, again the image in the different distortion situations is done the MOS subjective assessment, evaluation method is with reference to table 3.What three-dimensional display used is " 3DWINDOWS-19A01 " type Computerized 3 D imaging device that Tianjin Three-Dimensional Imaging Technology Co., Ltd. produces, as shown in Figure 4.The specification of three-dimensional form is as shown in table 4, and the main configuration requirement of computing machine is as shown in table 5.Test is finished by 15 observers, and wherein 3 is the professional person who is engaged in stereoscopic imaging technology research in the laboratory, and other staff are the layman.
In the following steps of embodiment, make X, Y represents respectively original image and distorted image, to each width of cloth distorted image, compares with original image respectively.The below describes as an example of double vision point stereo-picture teddy (as shown in Figure 5) example, and Fig. 6 has listed 7 kinds of distortion situations of single width viewpoint of the stereo-picture of teddy by name.Experiment adopts the part instrument among MATLAB and the photoshop to obtain distorted image as the realization means.
The first step is carried out respectively the phase reducing to the viewpoint of original image X to the viewpoint of (L1, R1) and distorted image Y to (L2, R2), obtains absolute difference image Diff (X), Diff (Y); The antipode figure of double vision point stereo-picture teddy as shown in Figure 7.
Second step utilizes the different image calculation weighted average construction of absolute parallax similarity WMSSIM, is designated as 3D_SSIM, and concrete steps are as follows:
1. the antipode image with stereo-picture carries out equal-sized division, is divided into B sub-block, and each sub-block is designated as B
i, pixel count is N in each sub-block
Sum
2. pass through formula
Calculate each sub-block B
iLuminance weights factor of influence v
i, grain details factor of influence t
i, locus factor of influence p
iWherein, I
Imax=max{I
1, I
2..., I
NBe the maximum brightness value in i the sub-block district, I
kBe brightness value,
Be the average brightness value in i sub-block district, x
Io, y
IoThe center of expression piecemeal i, x
c, y
cThe centre coordinate of expression original image, P represents that each point is to the ultimate range of centre coordinate in the original image.
3. pass through formula
Determine the weight coefficient w of each sub-block
iWherein, W
iThe factor of influence of each sub-block after comprehensive, W
SumThe combined influence factor W of all sub-blocks
iSummation, B is sub-block sum.
4. pass through formula
Calculate each sub-block B
iWeighted average construction similarity evaluation value.
5. distinguish the WMSSIM (X, Y) of calculated difference figure RGB component, obtain WMSSIM
R(X, Y), WMSSIM
G(X, Y), WMSSIM
B(X, Y) is by formula 3D_SSIM=E (WMSSIM
R(X, Y)+WMSSIM
G(X, Y)+WMSSIM
B(X, Y)) obtain parameter 3D_SSIM after the statistical average.
In the 3rd step, adopt the parallax that extracts stereo-picture based on the Stereo Matching Algorithm of image segmentation.Algorithm is realized block diagram as shown in Figure 3.Fig. 8 (a) is the disparity map of the original stereo-picture teddy that obtains by above-mentioned algorithm, Fig. 8 (b) is the disparity map of the original image teddy of standard, find after the contrast that disparity map with this algorithm extraction seems smoothly, edge's error is less, more accurate.
The 4th step is by formula PSNR
1=(PSNR
R+ PSNR
G+ PSNR
BThe Y-PSNR value PSNR of the error image Diff (X) of the original stereographic map of)/3 calculating and the error image Diff (Y) of distortion stereographic map
1PSNR
R, PSNR
G, PSNR
BBe respectively the PSNR value of three primary color components.
The 5th step is by formula PSNR
2=(PPSNR
R+ PPSNR
G+ PPSNR
BThe Y-PSNR value PSNR between original stereo-picture X and the distortion stereo-picture Y is calculated in)/3
2Wherein, PPSNR
R, PPSNR
G, PPSNR
BBe respectively the PPSNR value of three primary color components.
The 6th step, the initial evaluation value S_VALUE of calculating stereo-picture distortion level, computing formula is: S_VALUE=f
Normalization(PSNR
1)+f
Normalization(PSNR
2)+f
Normalization(3D_SSIM)).
The 7th step, after the parallax information of the parallax information of stereo-picture after the distortion and original stereo-picture extracted, calculating Y-PSNR DPSNR between the two.
In the 8th step, according to the distortion level of the size of DPSNR value judgement stereo-picture,, obtain the distorted image of different brackets, and then distribute to correction factor corresponding to each grade according to table 2 according to table 1 classification according to large young pathbreaker's distorted image of DPSNR value.With correction factor initial evaluation value S_VALUE is revised, finally obtained the quality assessment value 3D_VAL that calculates by this algorithm.
Fig. 9 is the evaluating objective quality value curve of stereo-picture in 12 kinds of distortion situations, is respectively teddy, baby3, Art, the Dolls 3D_VAL value in different distortion situations, and in order to make things convenient for observation and comparison, the design has done normalized with the objective evaluation value.Horizontal ordinate represents the stereo-picture in 12 kinds of distortion situations to be assessed among Fig. 9, and ordinate is the objective evaluation value of the image that obtains with this algorithm.Observation experiment found that the stereo-picture of Four types (different content) is similar with the curve tendency of the evaluation index that this algorithm calculates.Table 6 has provided the evaluation of estimate 3D_VAL in the different distortion situations of 7 points that the teddy image is corresponding after the emulation.
For the performance of verification system more intuitively, according to the subjective quality assessment stage division that table 3 provides the image in the different distortion situations is marked.For the consistance of subjective evaluation value is described, the design has drawn the scatter diagram of objective evaluation value and the subjective assessment value of distorted image to be evaluated, judges conforming quality by scatter diagram along whether concentrating of diagonal line distribution.
Figure 10 is the scatter diagram between 3D_VAL value and the corresponding DMOS value of 270 stereo-pictures to be evaluated.Figure 11 is that 78 stereo-picture traditional objective to be evaluated are estimated the scatter diagram between PSNR values and the corresponding subjective DMOS value, and the pure mathematics PSNR value of adding up is equivalent to the PSNR in the algorithm here
2Observing two width of cloth scatter diagrams can find out, 3D_VAL value loose along the lower left corner to the diagonal in the upper right corner, and relatively the concentrated area is distributed in around the diagonal line, shows that the 3D_VAL value is consistent substantially with the MOS value; And the evaluation of estimate PSNR that classic method obtains
2Scatter plot distributions then relatively disperse, section's spaced point is arranged even away from diagonal line.Related coefficient has also further been asked in the design experiment, for teddy figure, and the evaluation of estimate 3D_VAL that obtains by this algorithm and the coefficient R of DMOS value=0.8506, and the evaluation of estimate PSNR that obtains with pure PSNR method
2With the coefficient R of DMOS value=0.6885.
Can be drawn to draw a conclusion by above analysis: because simple PSNR algorithm is not considered the feature of stereo-picture, do not introduce the visual characteristic of human eye yet, therefore lower with the consistance of subjective assessment; And the design's algorithm has effectively utilized the feature of stereo-picture and considered human-eye visual characteristic role when observing image, so evaluation result is more consistent with subjective feeling.
Table 1 image fault grading
Grade | 1 | 2 | 3 | 4 | 5 |
DPSNR | <10 | [10,12] | (12,15) | [15,20] | >20 |
The correction factor of table 2 different brackets
Grade | 1 | 2 | 3 | 4 | 5 |
Correction factor | 0.5 | 0.4 | 0.28 | 0.22 | 0.2 |
Table 3 stereo image quality subjective assessment standards of grading
The three-dimensional form specification of table 4
Parameter | |
Model | The |
3D-19A01 type | |
Type | TFT |
The form diagonal line | 19 |
Aspect ration | |
4∶3 | |
Resolution | 1280×1024 |
Input signal | Digital signal input DVI-1, DVI-2 |
The power supply input | 100-240V~50/60HZ |
Environment for use | Humidity: 10%~80%; Temperature: 5~35 degree |
Purposes | Stereo display |
The main configuration requirement of table 5 computing machine
Parameter | Specification |
Operating system | Windows XP system |
Internal memory | 512M |
Video card | NVIDIA Quadro Fx supports 3dstereo to drive |
The stereo image quality objective evaluation value of the different distortions of table 6
Claims (3)
1. the stereo image quality evaluation method based on PSNR and SSIM is characterized in that, comprises the following steps:
The first step, with the viewpoint of original image X to (L1, R1) and the viewpoint of distorted image Y to (L2, R2) carry out respectively the phase reducing, obtain its absolute difference image Diff (X), Diff (Y), wherein, Diff (X)=| L1-R1|, Diff (Y)=| L2-R2|;
Second step utilizes the different image calculation weighted average construction of absolute parallax similarity WMSSIM, is designated as 3D_SSIM;
In the 3rd step, adopt the parallax that extracts stereo-picture based on the Stereo Matching Algorithm of image segmentation;
The 4th step is by formula PSNR
1=(PSNR
R+ PSNR
G+ PSNR
BThe Y-PSNR value PSNR of the error image Diff (X) of the original stereographic map of)/3 calculating and the error image Diff (Y) of distortion stereographic map
1, PSNR
R, PSNR
G, PSNR
BBe respectively the PSNR value of three primary color components;
The 5th step is by formula PSNR
2=(PPSNR
R+ PPSNR
G+ PPSNR
BThe Y-PSNR value PSNR between original stereo-picture X and the distortion stereo-picture Y is calculated in)/3
2, wherein, PPSNR
R, PPSNR
G, PPSNR
BBe respectively the PPSNR value of three primary color components;
The 6th step, the initial evaluation value S_VALUE of calculating stereo-picture distortion level, computing formula is:
S_VALUE=f
Normalization(PSNR
1)+f
Normalization(PSNR
2)+f
Normalization(3D_SSIM));
The 7th step, after the parallax information of the parallax information of stereo-picture after the distortion and original stereo-picture extracted, calculating Y-PSNR DPSNR between the two;
The 8th step, judge the distortion level of stereo-picture according to the size of DPSNR value, according to large young pathbreaker's distorted image of DPSNR value according to table 1 classification, obtain the distorted image of different brackets, and then distribute to correction factor corresponding to each grade according to table 2, with correction factor initial evaluation value S_VALUE is revised, finally obtain quality assessment value 3D_VAL;
Table 1 image fault grading
The correction factor of table 2 different brackets
In the 9th step, the stereoscopic image quality is estimated: the 3D_VAL value is larger, and the quality of image is better.
2. the stereo image quality evaluation method based on PSNR and SSIM as claimed in claim 1 is characterized in that the second step concrete steps are as follows:
1. the antipode image with stereo-picture carries out equal-sized division, is divided into B sub-block, and each sub-block is designated as B
i, pixel count is N in each sub-block
Sum
2. pass through formula
Calculate each sub-block B
iLuminance weights factor of influence v
i, grain details factor of influence t
i, locus factor of influence p
i, wherein, I
Imax=max{I
1, I
2..., I
NBe the maximum brightness value in i the sub-block district, I
kBe brightness value,
Be the average brightness value in i sub-block district, x
Io, y
IoThe center of expression piecemeal i, x
c, y
cThe centre coordinate of expression original image, P represents that each point is to the ultimate range of centre coordinate in the original image;
3. pass through formula
Determine the weight coefficient w of each sub-block
i, wherein, W
iThe factor of influence of each sub-block after comprehensive, W
SumThe combined influence factor W of all sub-blocks
iSummation, B is sub-block sum;
4. pass through formula
Calculate each sub-block B
iWeighted average construction similarity evaluation value;
5. distinguish R, G, the weighting sdffddsdfd average structure similarity value of B component: the WMSSIM of calculated difference image
R(X, Y), WMSSIM
G(X, Y), WMSSIM
B(X, Y) is by obtaining parameter 3D_SSIM, 3D_SSIM=E (WMSSIM after the formula statistical average
R(X, Y)+WMSSIM
G(X, Y)+WMSSIM
B(X, Y)).
3. the stereo image quality evaluation method based on PSNR and SSIM as claimed in claim 1 is characterized in that, the 3rd step,
Concrete steps are followed successively by: color segmentation, Window match, the extraction of parallax plane, parallax plane optimizing, parallax plane generate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012104311231A CN102982535A (en) | 2012-11-02 | 2012-11-02 | Stereo image quality evaluation method based on peak signal to noise ratio (PSNR) and structural similarity (SSIM) |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012104311231A CN102982535A (en) | 2012-11-02 | 2012-11-02 | Stereo image quality evaluation method based on peak signal to noise ratio (PSNR) and structural similarity (SSIM) |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102982535A true CN102982535A (en) | 2013-03-20 |
Family
ID=47856477
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2012104311231A Pending CN102982535A (en) | 2012-11-02 | 2012-11-02 | Stereo image quality evaluation method based on peak signal to noise ratio (PSNR) and structural similarity (SSIM) |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102982535A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104851098A (en) * | 2015-05-22 | 2015-08-19 | 天津大学 | Objective evaluation method for quality of three-dimensional image based on improved structural similarity |
CN104866864A (en) * | 2015-05-07 | 2015-08-26 | 天津大学 | Extreme learning machine for three-dimensional image quality objective evaluation |
CN104867138A (en) * | 2015-05-07 | 2015-08-26 | 天津大学 | Principal component analysis (PCA) and genetic algorithm (GA)-extreme learning machine (ELM)-based three-dimensional image quality objective evaluation method |
CN105976351A (en) * | 2016-03-31 | 2016-09-28 | 天津大学 | Central offset based three-dimensional image quality evaluation method |
CN106558047A (en) * | 2016-11-10 | 2017-04-05 | 复旦大学 | Color image quality evaluation method based on complementary colours small echo |
CN107038700A (en) * | 2016-11-24 | 2017-08-11 | 西安科技大学 | Soil moisture spatial distribution characteristic quantitative analysis method based on GSSIM |
CN107680068A (en) * | 2017-08-31 | 2018-02-09 | 田其冲 | A kind of digital image enhancement method for considering image naturalness |
CN109598299A (en) * | 2018-11-29 | 2019-04-09 | 微梦创科网络科技(中国)有限公司 | A kind of image similarity determines method, apparatus and electronic equipment |
CN109978933A (en) * | 2019-01-03 | 2019-07-05 | 北京中科慧眼科技有限公司 | The confidence level detection method of parallax information data, device and automated driving system |
CN110119731A (en) * | 2018-02-05 | 2019-08-13 | 广达电脑股份有限公司 | The device and method of 3-D image processing |
CN110208248A (en) * | 2019-06-28 | 2019-09-06 | 南京林业大学 | A method of identification Raman spectrum exception measuring signal |
CN111079787A (en) * | 2019-11-15 | 2020-04-28 | 上海航天控制技术研究所 | Position fingerprint identification method |
CN111583315A (en) * | 2020-04-23 | 2020-08-25 | 武汉卓目科技有限公司 | Novel visible light image and infrared image registration method and device |
CN111858345A (en) * | 2020-07-23 | 2020-10-30 | 深圳慕智科技有限公司 | Image sample generation capability multi-dimensional evaluation method based on antagonistic sample definition |
CN112446879A (en) * | 2021-01-06 | 2021-03-05 | 天津科技大学 | Contrast distortion image quality evaluation method based on image entropy |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080025400A1 (en) * | 2006-07-31 | 2008-01-31 | Kddi Corporation | Objective perceptual video quality evaluation apparatus |
CN101651769A (en) * | 2009-09-11 | 2010-02-17 | 南京大学 | Method for deciding color number and representative color value based on picture quality limiting condition |
-
2012
- 2012-11-02 CN CN2012104311231A patent/CN102982535A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080025400A1 (en) * | 2006-07-31 | 2008-01-31 | Kddi Corporation | Objective perceptual video quality evaluation apparatus |
CN101651769A (en) * | 2009-09-11 | 2010-02-17 | 南京大学 | Method for deciding color number and representative color value based on picture quality limiting condition |
Non-Patent Citations (1)
Title |
---|
沈丽丽: "立体视觉信息客观质量评价算法研究", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104866864A (en) * | 2015-05-07 | 2015-08-26 | 天津大学 | Extreme learning machine for three-dimensional image quality objective evaluation |
CN104867138A (en) * | 2015-05-07 | 2015-08-26 | 天津大学 | Principal component analysis (PCA) and genetic algorithm (GA)-extreme learning machine (ELM)-based three-dimensional image quality objective evaluation method |
CN104851098A (en) * | 2015-05-22 | 2015-08-19 | 天津大学 | Objective evaluation method for quality of three-dimensional image based on improved structural similarity |
CN105976351A (en) * | 2016-03-31 | 2016-09-28 | 天津大学 | Central offset based three-dimensional image quality evaluation method |
CN105976351B (en) * | 2016-03-31 | 2019-04-12 | 天津大学 | Stereo image quality evaluation method based on central offset |
CN106558047A (en) * | 2016-11-10 | 2017-04-05 | 复旦大学 | Color image quality evaluation method based on complementary colours small echo |
CN106558047B (en) * | 2016-11-10 | 2020-05-26 | 复旦大学 | Color image quality evaluation method based on complementary color wavelet |
CN107038700A (en) * | 2016-11-24 | 2017-08-11 | 西安科技大学 | Soil moisture spatial distribution characteristic quantitative analysis method based on GSSIM |
CN107038700B (en) * | 2016-11-24 | 2020-11-06 | 西安科技大学 | Soil humidity spatial distribution characteristic quantitative analysis method based on GSSIM |
CN107680068A (en) * | 2017-08-31 | 2018-02-09 | 田其冲 | A kind of digital image enhancement method for considering image naturalness |
CN110119731A (en) * | 2018-02-05 | 2019-08-13 | 广达电脑股份有限公司 | The device and method of 3-D image processing |
CN110119731B (en) * | 2018-02-05 | 2022-11-15 | 广达电脑股份有限公司 | Three-dimensional image processing device and method |
CN109598299A (en) * | 2018-11-29 | 2019-04-09 | 微梦创科网络科技(中国)有限公司 | A kind of image similarity determines method, apparatus and electronic equipment |
CN109978933A (en) * | 2019-01-03 | 2019-07-05 | 北京中科慧眼科技有限公司 | The confidence level detection method of parallax information data, device and automated driving system |
CN110208248A (en) * | 2019-06-28 | 2019-09-06 | 南京林业大学 | A method of identification Raman spectrum exception measuring signal |
CN110208248B (en) * | 2019-06-28 | 2021-11-19 | 南京林业大学 | Method for identifying abnormal measurement signal of Raman spectrum |
CN111079787A (en) * | 2019-11-15 | 2020-04-28 | 上海航天控制技术研究所 | Position fingerprint identification method |
CN111079787B (en) * | 2019-11-15 | 2023-06-06 | 上海航天控制技术研究所 | Position fingerprint identification method |
CN111583315A (en) * | 2020-04-23 | 2020-08-25 | 武汉卓目科技有限公司 | Novel visible light image and infrared image registration method and device |
CN111858345A (en) * | 2020-07-23 | 2020-10-30 | 深圳慕智科技有限公司 | Image sample generation capability multi-dimensional evaluation method based on antagonistic sample definition |
CN112446879A (en) * | 2021-01-06 | 2021-03-05 | 天津科技大学 | Contrast distortion image quality evaluation method based on image entropy |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102982535A (en) | Stereo image quality evaluation method based on peak signal to noise ratio (PSNR) and structural similarity (SSIM) | |
CN105407349B (en) | No-reference objective three-dimensional image quality evaluation method based on binocular visual perception | |
CN103581661B (en) | Method for evaluating visual comfort degree of three-dimensional image | |
CN102333233B (en) | Stereo image quality objective evaluation method based on visual perception | |
CN102170581B (en) | Human-visual-system (HVS)-based structural similarity (SSIM) and characteristic matching three-dimensional image quality evaluation method | |
CN104394403B (en) | A kind of stereoscopic video quality method for objectively evaluating towards compression artefacts | |
CN102663747B (en) | Stereo image objectivity quality evaluation method based on visual perception | |
CN106097327B (en) | In conjunction with the objective evaluation method for quality of stereo images of manifold feature and binocular characteristic | |
CN104811691B (en) | A kind of stereoscopic video quality method for objectively evaluating based on wavelet transformation | |
CN105338343B (en) | It is a kind of based on binocular perceive without refer to stereo image quality evaluation method | |
CN101976444B (en) | Pixel type based objective assessment method of image quality by utilizing structural similarity | |
CN101610425B (en) | Method for evaluating stereo image quality and device | |
CN105744256A (en) | Three-dimensional image quality objective evaluation method based on graph-based visual saliency | |
CN101872479A (en) | Three-dimensional image objective quality evaluation method | |
CN107481236A (en) | A kind of quality evaluating method of screen picture | |
CN104811693B (en) | A kind of stereo image vision comfort level method for objectively evaluating | |
CN103475897B (en) | Adaptive image quality evaluation method based on distortion type judgment | |
CN105654465B (en) | A kind of stereo image quality evaluation method filtered between the viewpoint using parallax compensation | |
CN102521825B (en) | Three-dimensional image quality objective evaluation method based on zero watermark | |
CN103945217B (en) | Based on complex wavelet domain half-blindness image quality evaluating method and the system of entropy | |
CN105049838A (en) | Objective evaluation method for compressing stereoscopic video quality | |
CN102708568B (en) | Stereoscopic image objective quality evaluation method on basis of structural distortion | |
CN109523513A (en) | Based on the sparse stereo image quality evaluation method for rebuilding color fusion image | |
CN103136748A (en) | Stereo-image quality objective evaluation method based on characteristic image | |
CN106023214B (en) | Image quality evaluating method and system based on central fovea view gradient-structure similitude |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130320 |