Research On Stereoscopic Image Quality Assessment Method In Three-dimensional Video Systems

Three-dimensional video system is a new multimedia format that can offer stereoscopic senseand interactive features, it also can provide the true feeling of immersive, the comprehensivelearning environment and high quality of visual enjoyment for human. With the development ofnetwork technologies and the increase of people’s material life, its application fields havewidespread deepened into social sectors, such as consumption entertainment, education,advertising, military aviation and so on. However, the segments of system are subjected by variousalgorithms or external constraints, it will inevitable affect the finally quality of the watching video.Consequently, it is one of focused researches to establish the reasonable and effective stereoscopicimage quality assessment method in the urgently resolved, because it is not only able to quantifythe degree of distortion in the three-dimensional video system, and measure the performance thealgorithms, but also can optimization the various algorithms. In addition, the human eye is mainreceiver of the video and image. So the thesis is based on the binocular visual characteristics tostudy stereoscopic image quality assessment in the three-dimensional video system, the researchesare as follows:(1)There are some problems of the depth accuracy and insufficiently considering thecharacteristics of human visual, for combining two-dimensional image quality assessment methodsand some depth information in current study. So a stereoscopic image objective quality assessmentmethod based on human visual systems is proposed, this method stands around the qualitysynthetic assessment of left and right perspectives model and the stereoscopic perceptionassessment model, the quality synthetic assessment of left and right perspective model considersthe nonlinear characteristics of brightness in human visual, and the characteristics of the differentregions having different structural features. The stereoscopic perception assessment modelsimulates the human multi-channel features and each channel’s sensitivity; make use of the humanvision signal noise ratio to assess. The results show that the correlation coefficients (CC) of theproposed assessment method and the subjective perception is above0.92, the Spearman rank ordercorrelation coefficients (SROCC) is greater than0.93, and the mean square errors (RMSE) is lessthan6.5. The above mentioned criterions precisely stereoscopic objective model is consistent withhuman visual properties and is of availability.(2)Concerning the less of combining the human visual characteristics with the stereoscopicimage’s nature features, a stereoscopic image objective quality assessment method based onsingular value decomposition (SVD) is presented, this method mainly takes the correlationbetween the image pixels and the degree of structural in the stereoscopic image into account from the perspective of the human visual, it also contains two models, one is the quality syntheticassessment of left and right perspectives, we take use of singular value decomposition which canrepresent the degree of the image distortion to evaluate the quality synthetic assessment of imagesbased on human visual characteristics, and then fusion the left and right qualities by thecharacteristics of binocular superposition. The other model is stereoscopic perception assessment,due to strong stability of image’s singular value characteristics, we calculate the distance of thesingular values and structural characteristics similarity of the absolute difference maps, and utilizethe statistical value of the global error to evaluate stereoscopic perception assessment. Finally,combine two models to describe the stereoscopic image quality. Experimental results show that theCC of the proposed method is above0.93, and the RMSE is less than6.2, it embodies that theproposed method is perfect consistent with subjective visual properties.(3) As for the current study is not fulfilling the needs of the practical application and lessconsidering the binocular visual information, a stereoscopic image objective quality assessmentmethod based on binocular information processing is proposed. According to the physiological andpsychology characteristics of human eyes, the binocular visual information go through the simplecells and complex cells, and then enter into the high level of the visual cortex to process, such asbinocular competition, binocular fusion, binocular suppression and so on. This method containstwo models: the binocular fusion quality assessment model and binocular stereo visual perceptionassessment model. Experimental results show that the CC of the proposed method is above0.94,and the RMSE is less than5.75. Compared with the previous assessment methods, this method hasa better consistency with human visual properties.