Research On Depth Map Coding For Three Dimentional Video Based On 3D-HEVC

Three dimensional(3D)video can provide an immersive stereoscopic vision for the viewer,and it has a promising application prospect in film industry,home entertainment industry and so on.Due to the great amount of 3D video data,valid compression and transmission techniques become the demanding issues to tackle.Therefore,3D video coding has become a hot topic in both academia and industry.Currently,multiview video plus depth(MVD)is the most efficient 3D video format which includes multiview texture videos with the associated multiview depth maps.Therefore,with the MVD format,some new virtual views can be synthesized between the original input views and can contribute to a more comfortable and free stereoscopic vision perception.In February 2015,the Joint Collaborative Team on 3D Video Coding Extension Development(JCT-3V)issued the 3D extension of High Efficiency Video Coding(3D-HEVC)standard aimed at increasing the coding efficiency for MVD format.Since the characteristics of the depth map are quite different from those of texture video,efficient depth map coding techniques are required to maintain the reconstructed quality of the depth map and further maintain the quality of the synthesized view.3D-HEVC can achieve an excellent depth map coding performance,but in turn brings the high depth map coding complexity.Therefore,in order to fulfill the demand of both efficient network transmission and comfortable viewing experience,valid coding schemes which are adopted to encode the depth map for 3D video efficiently and with high quality are the new challenges in the 3D video processing field.This thesis gives an in-deep investigation on the fast depth map coding algorithms for 3D video with high quality as well.The major work and contributions are listed as follows:1.The view synthesis optimization scheme is investigated in deep.According to the relationship between the depth map and the synthesized view,view synthesis optimization which is brought into the depth map coding decision processes including mode decision and motion/disparity estimation is focused.The coding performance and the computing complexity of the view synthesis optimization scheme are also analyzed.Through the view synthesis optimization,the quality of the synthesized virtual view and the quality of the reconstructed depth map can both act on guiding the depth coding decision processes.As a result,the quality of the synthesized virtual view can be maintained as well as the quality of the reconstructed depth map,and it contributes to increasing the coding efficiency of the depth map.2.A fast depth map coding algorithm based on texture characteristics is proposed.In order to decrease the high coding complexity due to the pixel-by-pixel view synthesis process in the conventional view synthesis optimization technique,by analyzing the influences of both depth map distortion and texture video distortion on the quality of the synthesized view,the texture characteristics of the co-located texture coding block are utilized for the current depth coding block to selectively skip the unnecessary view synthesis optimization process.The skipping threshold is set based on Intra DC prediction features and statistical approach.Experimental results show that the proposed algorithm can remarkably reduce the coding complexity while maintain the quality of the synthesized view.They also reveal that the proposed algorithm can especially have advantages for those sequences whose texture contents are smooth.3.A segment-based synthesized view distortion model for depth map is proposed.Through the analysis of the skip rates for different sizes of coding blocks of the conventional synthesized view distortion change model,and the distortion distribution features in both depth map and texture video,a segment-based synthesized view distortion calculation algorithm is proposed.The proposed algorithm recognizes the zero distortion segment based on both of the coded depth information and the coded texture information.Then the synthesized view distortion is calculated flexibly for each non-zero distortion segment which means the original line-based calculation is replaced by the flexible segment-based calculation.Experimental results show the proposed algorithm can increase the depth map coding speed meanwhile maintain the reconstructed quality of the original depth map and the quality of the synthesized view.They also show that the proposed algorithm especially has advantage for the test sequences with fast moving objects and complex texture contents.4.A region-based inter-picture weighted prediction algorithm using texture motion information is proposed for coding depth maps.Firstly,the motion information from the corresponding texture frame is utilized to distinguish between z-axis motion region and the background region.The weighted prediction parameters are calculated by the pixel within the z-axis motion region.Furthermore,in order to maintain the coding consistency,special weighted prediction parameters are set to handle the background region without intensity change.Finally,the proposed algorithm is cooperated with the multiple reference frame framework so that blocks from different regions can use different weighted reference frames to do inter-picture prediction.Experimental results show the proposed algorithm is able to compensate the local luminance change in depth map due to object's z-axis motion,and can improve the quality of depth map and the synthesized view without significant increase in coding time.