Research On 3D Virtual Viewpoint Video Quality Assessment

FVV(Free Viewpoint Video)system provides more viewing angles for users,furthermore increases human-computer interaction.Due to the limited number of video views which transmitted between networks,the virtual viewpoint rendering technology has become an essential part of the system.At present,the virtual viewpoint is mainly rendered by the DIBR(Depth Image Based Rendering)technique.However,the rendering result of DIBR is imperfect.In FVV system,the multiple combinations of 3D(Three-Dimensional)video among the viewpoint switching process leads to various watching experiences.To optimize the rendering process,we explore the factors affecting the quality of the 3D virtual viewpoint in FVV system.The research can also be used for video coding,and eventually improve the performance of FVV system.Therefore,this paper deeply studies the 3D virtual view image and video quality evaluation.(1)A new 3D virtual viewpoint image quality assessment measure is proposed.Considering the human visual characteristics,we classify the test image into content,depth perception and binocular characteristics parts.Firstly,the reference and distorted images are decomposed by wavelet,and then weighted by contrast sensitivity function respectively.The content score is obtained by calculating the structural similarity.Secondly,phase similarity between images on a foveal area is used to get the depth perception score.Thirdly,the distortion area which human eye is perceived in the binocular matching region is detected and quantified to obtain the binocular characteristics score.Finally,the content score,depth perception score and binocular characteristics score are integrated into an overall score.Experimental results show that the proposed algorithm has good consistency with subjective perception.(2)This paper proposes a quality assessment method for 3D virtual view video in FVV systems.We first utilize time-space domain features to detect the distortion critical region perceived by the human eye,and quantize the distortion degree of the region to obtain the initial video quality score.Then,the test videos are classified according to the multi-combination characteristics of FVV.Finally,the initial video quality score is refined to obtain the final quality score according to the video class.The metric is validated by the comparison with the related objective algorithms.Experimental results show that the proposed measure is an effective assessment tool of 3D virtual view video.