Research On Multi-Hypothesis Motion-Compensated Prediction In Video Communication

In recent years, Remote Distributed Video Surveillance (RDVS) system has been widely used in power system, railway traction power supply system and railway power system. To meet the demands of the robustly transmission of the video signal in high interference environment and the reduction of storage space and network bandwidth, one of the key tasks for RDVS is to compress video data efficiently.Motion-Compensated Prediction (MCP) , adopted in most video coding standards, is one of the hard core in current hybrid codecs. With the rapid development of network communication and multimedia processing technology, an advanced MCP technique named Multi-Hypothesis Motion-Compensated Prediction (MHMCP ) has been proposed. The multiple motion-compensated signals (hypotheses) are properly synthesized to predict the original video signal better, which means the accuracy of MCP can be further increasing. Not only higher coding efficiency could be achieved but also enhanced error resiliency of the compressed video bitstream could be obtained as well as the influence of error propagation could be reduced simultaneously. Therefore the conflicts between coding efficiency and error resilience could be trailed off to some extent. Recently MHMCP has been becoming a focus in the video communication, and the theoretical research and application of MHMCP are in the early stage.The main research contents and results of this dissertation are:Firstly, on the basis of discussing the classical MCP technique in detail, a new Rate-Distortion Optimal (RDO) based MHMCP algorithm is proposed in this paper. The shortcomings of the original bi-directional prediction pattern in H.264/AVC, such as not supporting refined partition patterns, needing to encode multiple Motion Vectors (MVs) and references information, and etc., could be overcame to certain degree. While preserving the optimal coding mode, the multiple hypotheses are particularly optimized in the proposed algorithm by utilizing the Refined Multi-Directional Searching Mode (RMDSM) , Scaled Motion-Vector Searching Mode (SMVSM) and MH-Coefficients Adjusting Mode (MHCAM) . The simulation results show that both prediction effect and rate-distortion performance of the video coder are improved.Secondly, considering the characteristic of multiple reference frames and complicated partition modes in motion estimation of H.264/AVC, a new Multi-Mode based Adaptive Motion Estimation (MMAME) algorithm is proposed. The optimal reference frame and partition mode could be determined rapidly by analyzing the relative information of spatial and temporal neighbor MacroBlocks (MBs). Furthermore, the initial searching center and searching length are chosen adaptively. Some fast termination strategies such as Sum of Absolute Difference (SAD) fast computing, all-zero-block detecting and overlapped searching points eliminating etc. are also adopted in MMAME. The simulation results show that the computing complexity of motion estimation is considerably reduced, while the coding efficiency is almost the same.Thirdly, MHMCP based error resilience and control technique is studied. On the encoder end, an mode selection algorithm of error resilient coding is proposed, which is realized by calculating the end-to-end distortion with the error propagation model in condition of different Package Lost Rate(PLR) on error-prone channels. With this algorithm, the error propagation is efficiently suppressed and the subjective video quality is improved. On the decoder end, a multi-hypothesis based adaptive error conceal algorithm is proposed. The criterion of Grade Similarity Matching (GSM) is introduced to improve the performance of Boundary Matching (BM) algorithm, since BM might cause incorrect MVs if there are abundant textures in the damaged MB. Via using enhanced set of candidate MVs, all matching signals obtained by different matching criterion would be properly synthesized. In this method, the mismatch of the final concealed signal is effectively removed and the quality of the concealed image is improved.Finally, a TMS320DM642 based H.264/AVC real-time encoding algorithm is realized. The system architecture, design principles, and several optimizing methods are discussed in detail. The experimental results show that the proposed methods mentioned in this section are effective and feasible.The research results of this dissertation are useful to achieve effective compression and robust transmission of video signal for RDVS. At the same time, it has positive effects on the development of other video applications.