CN103400378A - Method for objectively evaluating quality of three-dimensional image based on visual characteristics of human eyes - Google Patents

Abstract

The invention discloses a method for objectively evaluating the quality of a three-dimensional image based on the visual characteristics of human eyes. The method comprises the following steps: first respectively providing the original images and the test images of a left viewpoint and a right viewpoint; obtaining the quality measurement of the left viewpoint by a human visual system model according to the original image and the test image of the left viewpoint, obtaining the quality measurement of the right viewpoint in the same way, and then carrying out weighing combination on the quality measurements of the left viewpoint and the right viewpoint to obtain a left image quality evaluation and a right image quality evaluation; obtaining an original absolute difference value image according to the original images of the left viewpoint and the right viewpoint, obtaining a test absolute difference value image in the same way, then obtaining the absolute difference value image similarity measurement of the original absolute difference value image and the test absolute difference value image, and further obtaining a three-dimensional perception evaluation; and substituting the left image quality evaluation, the right image quality evaluation and the three-dimensional perception evaluation into an equation to calculate so as to obtain a quality evaluation result of the three-dimensional image. According to the method, the visual characteristics of the human eyes are combined, and meanwhile the influence of three-dimensional perception on the final quality of the three-dimensional image is studied in the view of the three-dimensional image so that the relevance between an objective evaluation model and subjective perception is improved.

Description

Look the objective evaluation method for quality of stereo images of characteristic based on human eye

Technical field

The invention belongs to computer vision field, be specifically related to a kind of objective evaluation method for quality of stereo images of looking characteristic based on human eye.

Background technology

Image quality evaluation is the study hotspot of image processing field, picture quality is the important indicator of more various image processing algorithm performance qualities and optimization system parameter, therefore in fields such as image acquisition, compression coding, Internet Transmissions, sets up effective image quality evaluation mechanism and is significant.In the last few years, along with the development of image processing techniques, the research in this field has attracted researchist's extensive concern, and the image quality evaluation algorithm emerges in an endless stream, typical model has the image quality evaluation model based on human visual system (HVS, Human Vision System).Along with greatly developing of multimedia technology and Internet technology, the three-dimensional video-frequency technology just develops rapidly.Compare with traditional media, when three-dimensional video-frequency can be created real scene impression more directly perceived for masses, need many one times at least of data to be processed.But when improving compression coding efficiency, also to guarantee the subjective perception of stereo-picture.Therefore, estimate stereo image quality, and the foundation objective evaluation model consistent with subjective assessment seems particularly important.At present, compare the plane picture quality assessment, the research of stereoscopic image quality assessment both at home and abroad compares less, most of objective evaluation model is not in conjunction with human-eye visual characteristic, or be the evaluation map image quality, and do not estimate three-dimensional perception, final appraisal results are poor with the subjective perception correlativity.

Summary of the invention

The present invention one of is intended to solve the problems of the technologies described above at least to a certain extent or provides at least a kind of useful business to select.For this reason, the object of the invention is to propose a kind of objective evaluation method for quality of stereo images of looking characteristic based on human eye.

According to the embodiment of the present invention look the objective evaluation method for quality of stereo images of characteristic based on human eye, comprise the following steps: that S1. provides left viewpoint test pattern, right viewpoint test pattern, left viewpoint original image and right viewpoint original image; S2. according to described left viewpoint test pattern and left viewpoint original image, utilize human vision system model, obtain left viewpoint quality metric, and, according to described right viewpoint test pattern and right viewpoint original image, utilize human vision system model, obtain right viewpoint quality metric, then, with described left viewpoint quality degree and the weighted array of right viewpoint quality metric, obtain the left and right image quality evaluation; S3. obtain original absolute difference image according to described left viewpoint original image and right viewpoint original image, obtain testing absolute difference image according to described left viewpoint test pattern and right viewpoint test pattern, then obtain absolute difference figure measuring similarity according to described original absolute difference image and test absolute difference image, further obtain three-dimensional perception evaluation; And S4. obtains the stereo image quality evaluation result with the image quality evaluation of described left and right and three-dimensional perception evaluation substitution linear regression fit regression equation calculation.

In one embodiment of the invention, described according to described left/right viewpoint test pattern and left/right viewpoint original image, utilize human vision system model, obtain left/right viewpoint quality metric, further comprise: S21. carries out wavelet transformation to described left/right viewpoint original image, then extract eigenwert, carry out the visual sensitivity weighting; S22. described left/right viewpoint test pattern is carried out wavelet transformation, then extract eigenwert, carry out the visual sensitivity weighting; And S23. utilizes canberra distance to measure to the result of described step S21 and step S22, obtains described left/right viewpoint quality metric Q _L, Q _R

In one embodiment of the invention, described with described left viewpoint quality degree and the weighted array of right viewpoint quality metric, adopt the weighted mean combination, i.e. left and right image quality evaluation Q ₁=0.5Q _L+ 0.5Q _R

In one embodiment of the invention, described step S3 further comprises: S31. is according to described left viewpoint original image L _orgWith right viewpoint original image R _orgObtain original absolute difference image X _org, X wherein _org=| R _org-L _org|, and, according to described left viewpoint test pattern L _disWith right viewpoint test pattern R _disObtain testing absolute difference image X _dis, X wherein _dis=| R _dis-L _dis|; S32. according to described original absolute difference image X _orgWith test absolute difference image X _dis, adopt the structural similarity of classic algorithm Wang to calculate described absolute difference figure measuring similarity SSIM according to following formula:

Wherein the size of piece is 8 * 8, u _x, u _yRepresent original and the average distorted image piece,

σ _xyRepresent original and the variance and covariance distorted image piece; And the average of the described SSIM value of S33. calculating, obtain described three-dimensional perception and estimate Q ₂

In one embodiment of the invention, described linear regression fit regression equation is: Q=0.689f (Q ₁)+0.358g (Q ₂)-0.731, wherein Q represents the stereo image quality evaluation result, f (Q ₁) be described left and right image quality evaluation Q ₁Normalized function, f (Q ₁)=100.253Q ₁ ³-215.838Q ₁ ²+ 161.005Q ₁+ 2.262, g (Q ₂) be that Q is estimated in described three-dimensional perception ₂Normalized function, g (Q ₂The 34.070Q of)=- ₂ ³+ 9.081Q ₂ ²-11.127Q ₁+ 38.897.

The objective evaluation method for quality of stereo images of looking characteristic based on human eye of the embodiment of the present invention, in conjunction with human-eye visual characteristic, is studied the impact of stereo-picture neutral body perception on final stereo image quality simultaneously, improves the correlativity of objective evaluation model and subjective perception.

Additional aspect of the present invention and advantage part in the following description provide, and part will become obviously from the following description, or by practice of the present invention, recognize.

Description of drawings

Above-mentioned and/or additional aspect of the present invention and advantage are from obviously and easily understanding becoming the description of embodiment in conjunction with following accompanying drawing, wherein:

Fig. 1 is the schematic diagram of stereo image quality objective evaluation model of the present invention;

Fig. 2 be the embodiment of the present invention look the process flow diagram of the objective evaluation method for quality of stereo images of characteristic based on human eye;

Fig. 3 is visual sensitivity function (CSF) curve;

Fig. 4 is that the sub-band coefficients standard deviation changes schematic diagram;

Fig. 5 is the structural similarity figure (SSIM_MAP) under different subjective qualities;

Fig. 6 (a) is left and right image quality evaluation distribution of results figure, and Fig. 6 (b) is three-dimensional perception evaluation result distribution plan;

Fig. 7 is evaluation result and subjective scoring DMOS scatter diagram.

Embodiment

Below describe embodiments of the invention in detail, the example of described embodiment is shown in the drawings, and wherein same or similar label represents same or similar element or the element with identical or similar functions from start to finish.Be exemplary below by the embodiment that is described with reference to the drawings, be intended to for explaining the present invention, and can not be interpreted as limitation of the present invention.

in description of the invention, it will be appreciated that, term " " center ", " vertically ", " laterally ", " length ", " width ", " thickness ", " on ", D score, " front ", " afterwards ", " left side ", " right side ", " vertically ", " level ", " top ", " end " " interior ", " outward ", " clockwise ", orientation or the position relationship of indications such as " counterclockwise " are based on orientation shown in the drawings or position relationship, only the present invention for convenience of description and simplified characterization, rather than device or the element of indication or hint indication must have specific orientation, with specific orientation structure and operation, therefore can not be interpreted as limitation of the present invention.

In addition, term " first ", " second " only are used for describing purpose, and can not be interpreted as indication or hint relative importance or the implicit quantity that indicates indicated technical characterictic.Thus, one or more these features can be expressed or impliedly be comprised to the feature that is limited with " first ", " second ".In description of the invention, the implication of " a plurality of " is two or more, unless otherwise expressly limited specifically.

In the present invention, unless otherwise clearly defined and limited, broad understanding should be done in the terms such as term " installation ", " being connected ", " connection ", " fixing ", for example, can be to be fixedly connected with, and can be also to removably connect, or connect integratedly; Can be mechanical connection, can be also to be electrically connected to; Can be directly to be connected, also can indirectly be connected by intermediary, can be the connection of two element internals.For the ordinary skill in the art, can understand as the case may be above-mentioned term concrete meaning in the present invention.

In the present invention, unless otherwise clearly defined and limited, First Characteristic Second Characteristic it " on " or D score can comprise that the first and second features directly contact, can comprise that also the first and second features are not directly contacts but by the other feature contact between them.And, First Characteristic Second Characteristic " on ", " top " and " above " comprise First Characteristic directly over Second Characteristic and oblique upper, or only represent that the First Characteristic level height is higher than Second Characteristic.First Characteristic Second Characteristic " under ", " below " and " below " comprise First Characteristic under Second Characteristic and tiltedly, or only represent that the First Characteristic level height is less than Second Characteristic.

Human visual system (HVS, Human Visual System) is by certain extremely complicated information handling system that connects to form by the neurocyte of a large amount of forms, Various Functions.For a long time, observation by some visual phenomenon to human eye and in conjunction with the achievement in research of vision physiological psychology aspect, it is found that the human visual system has a lot of characteristics, human eye vision susceptibility, hyperchannel characteristic, masking effect and three-dimensional perception etc. are wherein arranged, introduce these vision perception characteristics and can improve the correlativity of evaluation model and subjective scoring in image quality evaluation.Stereo image quality proposed by the invention is estimated objective models and is divided two parts, comprise the evaluation of left and right picture quality and the evaluation of three-dimensional perception, evaluation result and subjective scoring Linear Quasi synthesize regression equation, for the final appraisal results of stereoscopic image quality, objective models totally realizes block diagram as shown in Figure 1.

As shown in Figure 2, according to the embodiment of the present invention look the objective evaluation method for quality of stereo images of characteristic based on human eye, comprise the following steps:

S1., left viewpoint test pattern, right viewpoint test pattern, left viewpoint original image and right viewpoint original image are provided.

S2. according to described left viewpoint test pattern and left viewpoint original image, utilize human vision system model, obtain left viewpoint quality metric, and, according to described right viewpoint test pattern and right viewpoint original image, utilize human vision system model, obtain right viewpoint quality metric, then, with described left viewpoint quality degree and the weighted array of right viewpoint quality metric, obtain the left and right image quality evaluation.

In one embodiment of the invention, the process of calculating left/right viewpoint quality metric is: at first, described left/right viewpoint original image is carried out wavelet transformation, then extract eigenwert, carry out the visual sensitivity weighting.Then, described left/right viewpoint test pattern is carried out wavelet transformation, then extract eigenwert, carry out the visual sensitivity weighting; Finally, to the result of two ones of fronts, utilize the canberra distance to measure, obtain described left/right viewpoint quality metric Q _L, Q _R

In one embodiment of the invention, described with described left viewpoint quality degree and the weighted array of right viewpoint quality metric, adopt the weighted mean combination, i.e. left and right image quality evaluation Q ₁=0.5Q _L+ 0.5Q _R

S3. obtain original absolute difference image according to described left viewpoint original image and right viewpoint original image, obtain testing absolute difference image according to described left viewpoint test pattern and right viewpoint test pattern, then obtain absolute difference figure measuring similarity according to described original absolute difference image and test absolute difference image, further obtain three-dimensional perception evaluation.

Particularly, at first, according to described left viewpoint original image L _orgWith right viewpoint original image R _orgObtain original absolute difference image X _org, X wherein _org=| R _org-L _org|, and, according to described left viewpoint test pattern L _disWith right viewpoint test pattern R _disObtain testing absolute difference image X _dis, X wherein _dis=| R _dis-L _dis|; Then, according to described original absolute difference image X _orgWith test absolute difference image X _dis, adopt the structural similarity of classic algorithm Wang to calculate described absolute difference figure measuring similarity SSIM(structural similarity according to following formula):

Wherein the size of piece is 8 * 8, u _x, u _yRepresent original and the average distorted image piece,

σ _xyRepresent original and the variance and covariance distorted image piece.Finally, calculate the average of described SSIM value, obtain described three-dimensional perception and estimate Q ₂

S4. substitution linear regression fit regression equation calculation is estimated in the image quality evaluation of described left and right and three-dimensional perception, obtained the stereo image quality evaluation result.

Wherein, the linear regression fit regression equation is: Q=0.689f (Q ₁)+0.358g (Q ₂)-0.731, wherein Q represents the stereo image quality evaluation result, f (Q ₁) be described left and right image quality evaluation Q ₁Normalized function, f (Q ₁)=100.253Q ₁ ³-215.838Q ₁ ²+ 161.005Q ₁+ 2.262, g (Q ₂) be that Q is estimated in described three-dimensional perception ₂Normalized function, g (Q ₂The 34.070Q of)=- ₂ ³+ 9.081Q ₂ ²-11.127Q ₁+ 38.897.

In sum, the present invention, in conjunction with human-eye visual characteristic, studies the impact of stereo-picture neutral body perception on final stereo image quality simultaneously, improves the correlativity of objective evaluation model and subjective perception.

, for making those skilled in the art understand better the present invention, below in conjunction with embodiment, be further elaborated.

1. left and right image quality evaluation

The key property of HVS model comprises that visual properties, hyperchannel, visual sensitivity band are logical, interaction and the visual psychology features of different excitations between shielding effect, hyperchannel.Wherein non-linear, hyperchannel, the CSF band is logical and the shielding effect characteristic research is more, and corresponding computation model is arranged.These characteristics are relevant with the processing of image information directly or indirectly, therefore, introduce the various visual characteristics of human eye in image quality evaluating method, thereby make the result of objective evaluation more meet people's subjective judgement.The visual sensitivity band of introducing in the evaluation of this patent left and right picture quality in stereo-picture in the HVS characteristic leads to and the hyperchannel characteristic, and the HVS part of properties is carried out modeling.The left and right image after 5 grades of wavelet transformations, is divided into 11 frequency bands and 4 directions, and each passage is weighted by visual sensitivity function (CSF, Contrast Sensitivity Function) value.Extract the quality of utilizing the Canberra distance to measure the left and right image after the eigenwert of original and test left and right image.Take left visual point image as example, right viewpoint quality assessment is processed equally.

Human eye distinguishes that the ability of image detail is called sharpness of vision or visual space resolution, it with human eye can differentiate between images on the inverse of minimal visual angle of adjacent 2 represent.A large amount of researchs show, the spatial frequency characteristic of vision is the principal element that affects sharpness of vision, usually with visual sensitivity function CSF, describe relation between human visual system and frequency information.The expression formula of the visual sensitivity function that different researchers provide is different, is the function of frequency but all research experiments all show the visual sensitivity function, and has the characteristic of bandpass filter.Visual sensitivity reduces gradually with the too high or too low of frequency, and the curve of approximation of visual sensitivity function as shown in Figure 3.Wherein transverse axis is the spatial frequency of image, and unit is cd (cycle per degree, all every degree), and the longitudinal axis is the CSF functional value, shows the relative amplitude of eye response.

As shown in Figure 3, solid line represents the CSF curve on the horizontal and vertical direction, represents with A_VH, as shown in formula (1).Dotted line represents the CSF curve on angular direction is represented with A_D, as shown in formula (3).To the frequency f on angular direction _dAs shown in formula (2), f in formula (2) _h, f _vRepresent respectively the frequency on the horizontal and vertical direction.

$A\_\mathrm{VH}\left(f\right)=2.6[0.0192+0.114f]e^{[-{\left(0.114f\right)}^{1.1}]}---\left(1\right)$

$f_d=\sqrt{{f_h}^2+{{\mathrm{<figure-callout\; id="2"\; label="f\; v"\; filenames="HDA00003552815200021.png"\; state="\{\{state\}\}">f</figure-callout>}}_v}^2}=\sqrt{2}f---\left(2\right)$

$A\_D\left(f\right)=2.6[0.0192+0.114\sqrt{2}f]e^{[-{\left(0.114f\sqrt{2}\right)}^{1.1}]}---\left(3\right)$

There is different susceptibility in the cell of visual cortex to different visual information or excitation.And by to target identification, cover with adaptive research and think: all these features are activated in people's vision system, are to process at different passages, Here it is early stage Multichannel Theory.And further studied and point out afterwards: be not to isolate each other between the hyperchannel of vision mechanism, but influence each other, to produce optimum visual, but the interaction mechanism between hyperchannel is still not clear.

Generally, image quality evaluating method all can under the condition of considering applied environment and computing environment, be set up human visual system's frequency selectivity passage.Its objective is the multi-resolution characteristics of approximate simulation human eye vision perception.Aspect the Multichannel Decomposition algorithm, some appraisement systems can adopt comparatively complicated decomposition algorithm, for example the Multichannel Decomposition model that adopts of Daly, Lubin.But most evaluation method all can adopt some simple decomposition algorithms, such as wavelet transformation and dct transform.This patent adopts wavelet transformation to reduce the computation complexity of evaluation algorithms, utilizes the multichannel effect specific implementation flow process in the anthropomorphic eye of WAVELET TRANSFORM MODULUS visual characteristic to be:

At first, original image and test pattern are carried out respectively two-dimensional wavelet transformation, the direction of decomposition is horizontal direction, vertical direction and to angular direction, and the progression of decomposition is 5 grades.Then, due to the non-linear bandpass characteristics of CSF, the wavelet coefficient of different spaces frequency band after wavelet decomposition is weighted, weighted value is the mean value of CSF curve in frequency band.In low frequency part, and lowest frequency also comprises the DC component of image, so the weight of wavelet transformation lowest frequency subband is set herein, is 1 due to most concentration of energy of image.For 5 grades of wavelet decomposition, whole frequency band division is 11, according to CSF family curve correspondence, gets 11 weighted values, and the Weight selected of each wavelet sub-band of all directions is as shown in table 1:

The weights of each wavelet sub-band of table 1 all directions

For every layer of weights coefficient calculations (take the ground floor wavelet coefficient as example), wherein w as shown in formula (4) (5) _hv, w _dRepresent respectively the horizontal vertical direction and to the weights on angular direction.

$w_{\mathrm{hv}}=\frac{{\&Integral;}_{0.25}^{0.5}2.6[0.0192+0.114f]e^{[-{\left(0.114f\right)}^{1.1}]}\mathrm{df}}{0.5-0.25}---\left(4\right)$

$w_d=\frac{{\&Integral;}_{0.25}^{0.5}2.6[0.0192+0.114\sqrt{2}f]e^{[-{\left(0.114f\sqrt{2}\right)}^{1.1}]}\mathrm{df}}{0.5-0.25}---\left(5\right)$

The statistical property such as average, standard deviation is important textural characteristics.Feature extraction step is extracted respectively the average, variance of each sub-band coefficients as textural characteristics, is used for the similarity degree between reference metric image and test pattern.As shown in Figure 4, the image of selection is the Art stereo-picture, and type of distortion is Gaussian Blur, and each sub-band coefficients standard deviation of the image after wavelet transformation is along with the variation of image subjective quality is dull trend.Make μ _m,h, σ _m,hBe average and the variance of m level horizontal direction wavelet sub-band coefficient, as shown in formula (6) (7).

${\mathrm{\&mu;}}_{m,h}=\frac{{\mathrm{\&Sigma;}}_{i,j}{C}_{m,h}(i,j)}{N}---\left(6\right)$

${\mathrm{\&sigma;}}_{m,h}=\sqrt{\frac{{\mathrm{\&Sigma;}}_{i,j}{(C_{m,h}(i,j)-{\mathrm{\&mu;}}_{m,h})}^2}{N-1}}---\left(7\right)$

C wherein _m,hThe sub-band coefficients of (i, j) expression m level horizontal direction, N represents the sum of m level horizontal direction sub-band coefficients.In like manner can obtain vertically and to average and the standard deviation μ of sub-band coefficients on angular direction _m,v, σ _m,v, μ _m,v, σ _m,vThe sub-band coefficients standard deviation changes schematic diagram can be with reference to figure 4.

Remaining mass tolerance adopts the Canberra distance, can effectively distinguish the quality between the less test pattern of difference, in the calculated mass process, this models coupling vision hyperchannel characteristic, utilize Contrast sensitivity function the weights of wavelet coefficient of three directions in every layer of horizontal vertical diagonal angle of definite wavelet field.Take horizontal direction as example, as shown in formula (8),

$Q_h=\frac{1}{M}\underset{m}{\mathrm{\&Sigma;}}\frac{w_{m,h}|{\mathrm{\&sigma;}}_{m,h}^{\mathrm{org}}-{\mathrm{\&sigma;}}_{m,h}^{\mathrm{pro}}|}{({\mathrm{\&sigma;}}_{m,h}^{\mathrm{org}}+{\mathrm{\&sigma;}}_{m,h}^{\mathrm{pro}})}---\left(8\right)$

$Q_L=Q_h^L+Q_v^L+Q_d^L---\left(9\right)$

Q ₁=0.5(Q _L+Q _R) （10）

W wherein _m,hRepresent m level horizontal dimension coefficients weights, M is the sub-band sum of horizontal direction.

The sub-band coefficients standard deviation of m level horizontal direction original image and test pattern, vertically and to the angular direction similarity manage, and left and right picture quality is Q _L, Q _RFinal stereo image quality evaluation result is Q ₁, i.e. the weighting of left and right image quality evaluation result, the default weighting coefficient is identical, is 0.5.Q ₁Be worth greatlyr, show that stereo image quality is poorer.

2. three-dimensional perception evaluation

Stereoscopic sensation refers to that we experience the ability of the degree of depth, and this is a kind of ability of distinguishing the relative distance between the object that obviously is subjected to displacement.Also be fine with the simple eye relative position that goes to experience between object, but it is a lateral shift, eyes can obtain the different impression of same object, i.e. different picture, and can obtain sharper stereoscopic sensation depth discrimination.Relief quality can have influence on stereo image quality.

Study the absolute difference figure that shows original left and right image and the absolute difference figure similarity of testing the left and right image and can estimate the three-dimensional perception of stereo-picture, absolute difference figure is more similar, and the stereo-picture stereoscopic sensation is stronger.

Obtain the absolute difference image of original and test left and right image as shown in formula (9) (10), wherein L _org, R _orgBe respectively the left and right image of original image, R _dis, L _disBe respectively the left and right image of test pattern, X is the absolute difference figure of left and right image, X _orgFor original left and right image absolute difference figure, X _disFor test left and right image absolute difference figure.

X _org=|R _org-L _org| （9）

X _dis=|R _dis-L _dis| （10）

The structural similarity of employing classic algorithm Wang is estimated the quality of absolute difference image, as shown in formula (11),

$\mathrm{SSIM}=\frac{({2u}_x{u}_y+C_1)(2{\mathrm{\&sigma;}}_{\mathrm{xy}}+C_2)}{(u_x^2+{\mathrm{<figure-callout\; id="2"\; label="u\; y"\; filenames="HDA00003552815200021.png"\; state="\{\{state\}\}">u</figure-callout>}}_y^{}+C_1)({\mathrm{\&sigma;}}_x^2+{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA00003552815200021.png"\; state="\{\{state\}\}">y</figure-callout>}}^2+C_2)}---\left(11\right)$

Wherein the size of piece is 8 * 8, u _x, u _yRepresent original and the average distorted image piece,

σ _xyRepresent original and the variance and covariance distorted image piece, three-dimensional perception evaluation result Q ₂Average for the SSIM value.Structural similarity figure (SSIM_MAP) as shown in Figure 5.Position explanation SSIM value more black in figure is less, and left and right viewpoint absolute difference is more dissimilar, and by seeing in Fig. 5, the quality of Fig. 5 (a) is better than Fig. 5 (b), with the subjective assessment result, is consistent.

3 stereo image quality objective evaluations

For obtaining the funtcional relationship between left and right picture quality and three-dimensional perceived quality and subjective assessment value, the stereographic map image set comprises 10 groups of high definitions (resolution is greater than 1330 * 1110) test pattern, contained as personage, static state, enriched the different characteristics of image such as texture, all are participated in the stereo-picture pair of test, left visual point image is not done any distortion and is processed, right visual point image is carried out 4 kinds of distortions process, comprise JPEG compression, JPEG2000 compression, white noise distortion, Gaussian Blur distortion.Every width reference picture is carried out in various degree processing, and image library gives average subjective scoring difference.Every a pair of image is calculated respectively left and right picture quality and three-dimensional perception evaluation result Q ₁And Q ₂, and investigate relation between two evaluation results and subjective assessment value DMOS.

Distribution situation according to left and right picture quality and three-dimensional perception in Fig. 6, the response function of both matches is f and g, as shown in formula (12) (13), in Fig. 6, horizontal ordinate represents that left and right image quality evaluation model and three-dimensional perception evaluation model normalize to the value of [0,1].

f(Q ₁)=100.253Q ₁ ³-215.838Q ₁ ²+161.005Q ₁+2.262 （12）

g(Q ₂)=-34.070Q ₂ ³+9.081Q ₂ ²-11.127Q ₁+38.897 （13）

As can be seen from Figure 6, the distribution of left and right image quality evaluation result and three-dimensional perception evaluation result all has regularity, be distributed in f and g curve near.Adopting the mode of linear regression that response function f and g are carried out match, is the evaluation result of stereoscopic image quality finally, can obtain regression equation Q=0.689f (Q ₁)+0.358g (Q ₂)-0.731.Q represents the final appraisal results of stereo image quality, f and g front coefficient reflect the importance in left and right image quality evaluation result and the evaluation of three-dimensional perception evaluation result stereo image quality to a certain extent, can draw both left and right picture qualities all relevant with the perceived quality of stereo-picture with three-dimensional perception.

Stereo-picture storehouse neutral body image has 380 pairs. and wherein original image is 10 pairs, and stereo-picture to be evaluated is 370 pairs.Image library gives average subjective scoring difference (DMOS, Difference Mean Opinion Scores), DMOS is the difference (DMOS=100-MOS) of subjective scoring average (MOS) and full marks (100), therefore, the larger presentation video quality of DMOS value is poorer, the less presentation video quality of DMOS value is better, and the span of DMOS is [0,100].The present invention utilizes two of the evaluate image quality evaluating method objective parameters commonly used as evaluation index: the related coefficient CC under the non-linear regression condition (Correlation Coefficient) and ROCC (Rank-Order Correlation Coefficient), the former reflects the accuracy of objective models, and the latter is reflected its monotonicity.The non-linear regression function adopts the Logistic function in document, and the higher explanation method for objectively evaluating of CC and ROCC value and DMOS correlativity are better.As shown in Figure 7, CC and the ROCC coefficient of reflection accuracy and monotonicity are as shown in table 2 for the scatter diagram of model evaluation result and subjective scoring.

The performance index of table 2 objective models

Fig. 7 in comparison diagram 7 (a) and Fig. 7 (b) two figure, before not adding the stereoscopic sensation evaluation, especially the image after JPEG compression and JPEG2000 compress, the scatter diagram of left and right image quality evaluation evaluation result and subjective scoring DMOS relatively disperses, and increase after three-dimensional perception evaluation, scatter diagram is more concentrated, and is more consistent with the correlativity of subjective assessment result, no matter be accuracy, or monotonicity has nearly all reached more than 0.9.

need to prove, describe and can be understood in process flow diagram of the present invention or in this any process of otherwise describing or method, expression comprises the module of code of the executable instruction of the step that one or more is used to realize specific logical function or process, fragment or part, and the scope of the preferred embodiment of the present invention comprises other realization, wherein can be not according to order shown or that discuss, comprise according to related function by the mode of basic while or by opposite order, carry out function, this should be understood by the embodiments of the invention person of ordinary skill in the field.

In the description of this instructions, the description of reference term " embodiment ", " some embodiment ", " example ", " concrete example " or " some examples " etc. means to be contained at least one embodiment of the present invention or example in conjunction with specific features, structure, material or the characteristics of this embodiment or example description.In this manual, the schematic statement of above-mentioned term not necessarily referred to identical embodiment or example.And the specific features of description, structure, material or characteristics can be with suitable mode combinations in any one or more embodiment or example.

Although the above has illustrated and has described embodiments of the invention, be understandable that, above-described embodiment is exemplary, can not be interpreted as limitation of the present invention, those of ordinary skill in the art is not in the situation that break away from principle of the present invention and aim can change above-described embodiment within the scope of the invention, modification, replacement and modification.

Claims (5)

1. an objective evaluation method for quality of stereo images of looking characteristic based on human eye, is characterized in that, comprises the following steps:

S1., left viewpoint test pattern, right viewpoint test pattern, left viewpoint original image and right viewpoint original image are provided;

S2. according to described left viewpoint test pattern and left viewpoint original image, utilize human vision system model, obtain left viewpoint quality metric, and, according to described right viewpoint test pattern and right viewpoint original image, utilize human vision system model, obtain right viewpoint quality metric, then, with described left viewpoint quality degree and the weighted array of right viewpoint quality metric, obtain the left and right image quality evaluation;

S3. obtain original absolute difference image according to described left viewpoint original image and right viewpoint original image, obtain testing absolute difference image according to described left viewpoint test pattern and right viewpoint test pattern, then obtain absolute difference figure measuring similarity according to described original absolute difference image and test absolute difference image, further obtain three-dimensional perception evaluation; And

S4. substitution linear regression fit regression equation calculation is estimated in the image quality evaluation of described left and right and three-dimensional perception, obtained the stereo image quality evaluation result.

2. the objective evaluation method for quality of stereo images of looking characteristic based on human eye as claimed in claim 1, it is characterized in that, described according to described left/right viewpoint test pattern and left/right viewpoint original image, utilize human vision system model, obtain left/right viewpoint quality metric, further comprise:

S21. described left/right viewpoint original image is carried out wavelet transformation, then extract eigenwert, carry out the visual sensitivity weighting;

S22. described left/right viewpoint test pattern is carried out wavelet transformation, then extract eigenwert, carry out the visual sensitivity weighting; And

S23. utilize the canberra distance to measure to the result of described step S21 and step S22, obtain described left/right viewpoint quality metric Q _L, Q _R

3. the objective evaluation method for quality of stereo images of looking characteristic based on human eye as claimed in claim 1, is characterized in that, and is described with described left viewpoint quality degree and the weighted array of right viewpoint quality metric, adopts the weighted mean combination, i.e. left and right image quality evaluation Q ₁=0.5Q _L+ 0.5Q _R

4. the objective evaluation method for quality of stereo images of looking characteristic based on human eye as claimed in claim 1, is characterized in that, described step S3 further comprises:

S31. according to described left viewpoint original image L _orgWith right viewpoint original image R _orgObtain original absolute difference image X _org, X wherein _org=| R _org-L _org|, and, according to described left viewpoint test pattern L _disWith right viewpoint test pattern R _disObtain testing absolute difference image X _dis, X wherein _dis=| R _dis-L _dis|;

S32. according to described original absolute difference image X _orgWith test absolute difference image X _dis, adopt classic algorithm Wang structural similarity to calculate described absolute difference figure measuring similarity SSIM according to following formula:

Wherein the size of piece is 8 * 8, u _x, u _yRepresent original and the average distorted image piece,

σ _xyRepresent original and the variance and covariance distorted image piece; And

S33. calculate the average of described SSIM value, obtain described three-dimensional perception and estimate Q ₂

5. the objective evaluation method for quality of stereo images of looking characteristic based on human eye as claimed in claim 1, is characterized in that, described linear regression fit regression equation is: Q=0.689f (Q ₁)+0.358g (Q ₂)-0.731, wherein Q represents the stereo image quality evaluation result, f (Q ₁) be described left and right image quality evaluation Q ₁Normalized function, f (Q ₁)=100.253Q ₁ ³-215.838Q ₁ ²+ 161.005Q ₁+ 2.262, g (Q ₂) be that Q is estimated in described three-dimensional perception ₂Normalized function, g (Q ₂The 34.070Q of)=- ₂ ³+ 9.081Q ₂ ²-11.127Q ₁+ 38.897.

Images