Research On Inter-Frame Error Concealment Techniques For Multing View Stereo Video

With the increasing demand of consumers for high definition three-dimensional video(HD-3DV),the Joint Collaborative Team on Video Coding(JCT-VC)proposed a multiview video coding standard,i.e.MV-HEVC and 3D-HEVC,for 3D video coding based on HEVC(High Efficiency Video Coding)standard in 2014.Compared with the traditional H.264 standard,the compression efficiency is increased by one time for 3D video under the condition of same decoding quality.This implies less compression bitrate,while also means that once packages loss or channel errors occur,the reconstructed video quality at decoder will be severely declined.Since multiview video has higher data correlation than single view video,thus,effective error control technique is indispensable to stereo/multiview video transmission.The multiview error concealment technique is to reconstruct the lost blocks by using the strong temporal,spatial and inter-view correlation of the decoded stereo/multiview video frames.The key of inter-frame error concealment of stereo/multiview video is how to accurately reconstruct the motion/disparity vectors of the missing coding units by using the temporal,spatial and inter-view correlations in stereo/multiview video,and how to use the binocular perception feature of human eyes to improve the reconstruction quality.Based on the analysis of core technology of MV-HEVC and the human visual perception model,this thesis focuses on the research of interview error concealment technique of MV-HEVC at decoder,and proposes a new error concealment algorithm for MV-HEVC standard.In summary,the main works of this thesis are listed as follows:1.We propose an efficient error concealment algorithm based on MV-HEVC.The algorithm classifies the damaged coding units into static blocks and motion blocks,where the static blocks are directly restored by using a direct copy,while the motion blocks are reconstructed by motion/disparity compensation–based error concealment.In the stage of motion/disparity compensation method,we first consider the strong correlation of coding depth between corrupted blocks and its neighboring blocks,and use the depths of neighboring blocks to adaptively split the damaged blocks.Then,we establish a motion/disparity vectors set for each damaged sub-blocks.Next,based on the limitation of the traditional matching criterion in the boundary matching of a CU block,we propose a new block matching criterion which considers both the pixel smoothness and the similarity of the texture structure.Finally,according to the value of CMAD,we choose to use the optimal motion/ disparity vector selected from the motion/disparity vectors set as a concealed motion/disparity vector to reconstruct the corrupted block,or use a motion/disparity vector interpolation method to obtain a new motion/disparity vector as the concealed vector to restore the lost block.Experimental results show that the proposed algorithm can improve the reconstruction accuracy of lost blocks,maintain a better texture consistency of the recovered unit and its neighboring units,and thus improve the reconstruction quality of the error frame.2.We propose a stereoscopic video error concealment algorithm based on binocular visual perception.A pioneering work in proposed algorithm is to introduce the binocular just noticeable difference(BJND)model into the error concealment.In the stage of CU partitioning,we not only consider the content characteristics of the video,but also adaptively split the damaged blocks via the BJND model and the motion/disparity vector of neighboring blocks.Moreover,a new matching criterion conbined with the BJND model,DBJND,is proposed by considering the human eye's allowable distortions of the left and right view.If the matching result by using the optimal motion/disparity vector selected from the constructed motion/disparity vectors set is larger than the DBJND,the current damaged block is replaced by searching for the best matching block in the predetermined search range at the inter-view frame.Experiment results show that the concealed frames restored by our algorithm not only achieves high PSNR,but also maintains a high consistency with the visual perceiption of human beings.