Research On 3D Video Coding Based On Virtual View Quality

With the development of 3D display and interactive multimedia systems,Multi-view Video plus Depth(MVD)video has become the main direction of the new generation 3D video application and attracted significant interests.MVD can provide the user with more natural and live visual experience in immersion and 3D sense compared with the traditional 2D video,which can also provide arbitrary viewpoint freely within a given range.MVD includes colorful texture images as well as gray depth maps.The depth map contains the scene geometric information and is utilized to synthesis the virtual views through a Depth Image Based Rendering(DIBR)technology instead of display.The depth map has very different characteristics compared with the 2D videos which makes the traditional coding tools cannot encode it efficiently and will induce virtual view distortion.Besides the coding efficiency of texture image and depth map,bit allocation between texture image and depth map also has a great effect on MVD video coding efficiency.Therefore,high efficiency MVD coding has become the research hotspot and attracted increasing researches inside and outside the country.The major work of this paper is as follows:Two depth map coding algorithms based on virtual view distortion are proposed to reduce the synthesized virtual view distortion induced by depth map compression.1)Based on analysing the statistics of different virtual view distortion induced by different depth map compression,an exponential model to estimate the virtual view distortion induced by depth map distortion is proposed.The model parameters are calculated through pre-processing of the 3D video,and the model is implemented on the Rate Distortion Optimization(RDO)of depth map coding;2)Based on the first algorithm,by analyzing the virtual view distortion induced by different depth map regions,a virtual view distortion estimation model which takes both texture image and depth map characteristics into account is proposed,and the model is implemented on to the RDO of depth map coding.We also adjust depth map coding Quantization Parameters(QPs)in different regions to improve the synthesized virtual view quality.Experimental results show that both the two methods can estimate the virtual view distortion induced by depth map compression accurately,improve the virtual view quality as well as depth map coding efficiency,and the coding complexity is reduced compared with the View Synthesis Optimization(VSO)technique in 3D-HEVC.The second algorithm does not need pre-processing thus is much faster compared with the first algorithm and is adaptive to different 3D videos.Based on the characteristic and utility of depth map,a fast depth map coding algorithm with high virtual view quality is proposed.The algorithm includes: 1)based on the characteristics of both texture image and depth map,the virtual view distortions of different depth map regions are estimated.For depth map regions which induce low virtual view distortion,the virtual view distortion estimation process is skipped in order to reduce the depth map coding complexity;2)encode depth map regions with low virtual view distortion in large Coding Unit(CU)size and terminate the quad-tree coding structure search in 3D-HEVC,and encode depth map regions with high virtual view distortion in smaller CU size,in order to reduce the high complexity of quad-tree coding structure partition in 3D-HEVC.Experimental results demonstrate that the proposed algorithm can reduce the depth map coding complexity significantly with negligible synthesized virtual view quality degradation.Besides the high coding efficiency of both texture image and depth map,bit allocation between texture image and depth map also has a great effect on 3D video coding.In order to improve 3D video coding efficiency,we propose a bit allocation algorithm which includes: 1)a more accurate relationship between virtual view distortion and QPs of texture image and depth map;2)based on the quality difference of texture image in adjacent views,bits are allocated between views adaptively at Group of Pictures(GOP)level to improve the synthesized virtual view quality;3)based on the synthesized virtual view distortion difference between the previous encoded two GOPs,the bit ratio between texture image and depth map of the current GOP is adjusted dynamically to improve the synthesized virtual view quality;4)bits are allocated between texture image and depth map based on adjusting the Lagrangian Multiplier(LM)of depth map.Experimental results show that the proposed algorithm can allocate bits between adjacent views and between texture image and depth map optimally.The algorithm can achieve best virtual view quality adaptively for different 3D videos with extremely low coding complexity,which can improve the practicability of MVD video.