| Multimedia systems which is based on the three-dimensional video technology can provide users with a sense of three-dimensional,so that the process of users experience can have exposure to the realism.However,in the process of image or video processing,the problem of distortion caused by the process of acquisition,coding,transmission,drawing and display is unavoidable.Therefore,it is necessary to study the related quality evaluation techniques to improve the performance of the system in image or video processing,and improve visual perception.This thesis study the influence of human visual perception on objective quality evaluation model.The main research contents are as follows:(1)From the perspective of stereoscopic perception,considering the characteristics of binocular perception,combined with the masking effect of binocular vision and the characteristics of retinal corresponding points,the related stereoscopic image quality evaluation model is established.Firstly,the binocular masking effect of the human eye is simulated by using the joint just noticeable difference(JJND),and the stereo image pair is adjusted.Then the left and right view image quality is evaluated.At the same time,according to the characteristics of the retina corresponding point of the same characteristics and the human eye-sensitive characteristics of the use of parallax to establish the relevant three-dimensional perception model.Finally,the left and right view quality is combined with the stereo perception quality to predict the stereo image quality.The method is consistent with the subjective perception performance,CC value is 0.9293,SROCC reaches 0.9339,RMSE is 5.8640.(2)From the perspective of color perception and space-time perception,an objective algorithm model for video quality evaluation is proposed.Considering the important role of color information in visual perception and the processing of multidimensional data,the color video frame is processed by tensor decomposition,the color is dimensioned and the new 3D video is obtained.At this time,the 3D Discrete wavelet transform(3D-DWT)can be applied to this 3D video to get time and space information.According to the characteristics of the three-dimensional wavelet signal,the contrast sensitivity weight obtained by the two-dimensional wavelet transform is extended to three dimensions by the principle of multiplication.The similarity of the three-dimensional structure similarity is measured and the final evaluation result is obtained by combining the asymmetry in the time domain.The proposed CC and SROCC values on both the LIVE and CSIQ video libraries have reached above 0.8.(3)Based on the motion perception of video from the human eye,a non-reference stereo video quality evaluation method based on time-varying information of tensor decompositiondomain is studied.In this paper,we find that the motion information of tensor domain of stereo video has statistical characteristics,The statistical distribution can distinguish the different distortions type and distortion degree.This feature contains longer time information,to a certain extent,can handle the problem that the short-term distortion can be ignored by the human eye but was treated as distorted.Motion information extraction feature,a non-reference stereo video quality evaluation method based on time-varying information in tensor decomposition domain is studied.Firstly,the color,time and viewpoint are dimensioned by tensor decomposition,and the motion information map of time-varying information in different viewpoints and motion information of time-varying information about different viewpoints.Then,the parameters of generalized Gaussian distribution(GGD)and the asymmetric generalized Gaussian distribution(AGGD)model,the spatial entropy,the spectral entropy and the stereo entropy are extracted from the motion image.Finally,the use of random forest fusion perception features to get the final video quality.The experiment is compared with the pixel domain at the same time,and the influence of the time dimension of the tensor on the performance of the algorithm is analyzed.The experimental results show that the CC value is 0.8611 and the SROCC value is 0.8394,which proves that the evaluation result of this model is consistent with the subjective perception. |
