Research On Error Concealment And Quality Improvement For Multi-view Video

With the progress of the times and the development in multimedia technology,consumers are no longer satisfied with the visual experience brought by traditional video.In order to meet people's requirements for better interactivity,multi-view technology has gradually come into people's sight.Compared to the 2D video,the multi-view video can generate parallax by using multiple views to give the viewers the feeling of depth,which is not provided by the 2D video.Moreover,multi-view video can provide users with various angles of view to freely choose viewing angles,compared to the traditional 3D video.However,the excellent viewing experience is accompanied by a vast amount of data of the multi-view video.Although the emerging coding scheme can provide high compression efficiency,the contradiction between the compression ratio and error robust will significantly reduce the decoding quality of multi-view video suffering packet loss,especially in the situation with high real-time requirements,such as video conferencing and telemedicine.When a packet loss occurs,the traditional error-resistance technology will face the problem of high complexity and poor real-time.Therefore,for the packet loss problem of multi-view video,it is of considerable significance to study a low-latency and effective error region reconstruction algorithm.For these reasons,this thesis mainly studies the multi-view video reconstruction problem under low latency requirements.The right view loss problem of stereoscopic video(two views)and the intermediate view loss problem of multi-view videos are analyzed.By combining the key technologies of MV-HEVC,the traditional error concealment algorithm is improved and innovated.Based on these,two efficient reconstruction algorithms for multi-view videos are proposed.The main contributions of this thesis are listed as follows:1.We propose a multi-view video error concealment algorithm based on disparity vector extrapolation.The algorithm can solve the right view loss problem of stereoscopic video under HBP structure and the non-independent view loss problem of multi-view video under HBP structure.Aiming to the problem of the traditional motion vector extrapolation algorithm cannot be used for the HBP structure,we propose a disparity vector extrapolation algorithm based on the disparity estimation.The proposed algorithm solves the applicability problem of the extrapolation algorithm caused by HBP structure by combining spatial-temporal and disparity information.And an effective hole interpolation algorithm is proposed for the hole that occurs in the traditional extrapolation algorithm.The experimental results show that compared with the conventional algorithm,the proposed algorithm has a significant improvement in objective quality evaluation and subjective quality evaluation.2.We propose a novel error concealment algorithm based on adaptive block partition and residual information.The algorithm can solve the middle-view loss problem of multi-view video under HBP structure.In order to solve the blocking artifact caused by the reconstruction of sub-blocks with fixed-sizes,the algorithm proposes an adaptive block partitioning method based on residual information.Aiming at the error propagation problem of the traditional boundary matching algorithm,we present a sub-block reconstruction sorting algorithm combining depth information,residual information,and prediction information.Based on these,an error concealment algorithm combining spatial-temporal and disparity information is proposed.The experimental results show that the proposed algorithm can reduce the blocking artifact and achieve significant improvement in objective quality and subjective quality,compared with the state-of-the-art algorithms.