Study On Depth Image Objective Quality Assessment

In recent years,virtual view synthesis has been increasingly popular due to the wide applications of multi-view and free-viewpoint videos.In view synthesis,texture images are rendered to generate the new viewpoint with the guidance of the depth images.The quality of depth images is vital for generating high-quality synthesized views.While the impact of texture image and the rendering process on the quality of the synthesized view have been extensively studied,the quality evaluation of depth image remains largely unexplored.The existing natural image quality assessment metrics cannot accurately predict the quality of depth images,so efficient depth image quality assessment method must be proposed.In view of the above problems,two methods are proposed: depth image quality assessment based on edge blocks structure similarity and quality index of depth images based on statistics of edge profiles.The main work and contributions of this thesis are as follows:1.A reduced-reference(RR)depth image quality assessment method based on edge blocks structure similarity is proposed.In view of the distortion characteristics of depth image,the depth image is divided into nonoverlapping blocks,and then block is judged as edge block if the ratio of the edge occupying the block is bigger than a certain value.Then a weighted combination score of the image edge block depth value and the gradient similarity result is computed as the score of the depth image block,and then the average score of all the edge blocks is computed as the overall quality score of the depth image.The experimental results on the MCL-3D dataset show that the proposed method can effectively predict the quality of depth images.2.A no-reference(NR)quality index for depth images by modeling the statistics of edge profiles in a multi-scale framework is proposed.The Canny operator is first utilized to locate the edges in depth images.Then the edge profiles are constructed,based on which the gradient magnitude(GM)and Laplace of Gaussian(LoG)are extracted as first-order and second-order statistical features for portraying the distortions in depth images.Then the Weibull distribution and the asymmetric generalized Gaussian distribution(AGGD)are used to fit the histogram distributions of the first and second order features respectively,and the parameters of the two distributions are used as the final features.Finally,the random forest regression is employed to build the quality model for depth images.Extensive experiments are conducted on two public view synthesis image/video quality databases.The experimental results and comparisons demonstrate that the proposed metric outperforms the relevant state-of-the-art quality metrics.Further,it is also advantageous in terms of the generalization ability.