Researches On Video Coding Techniques For Transmission Over The Internet

With the success of both the Internet and video coding techniques, videotransmission over the Internet has become a hotspot in recent years. Video streamtypically has the requirements of wider bandwidth, less delay and higher reliability.However, these requirements can not be adequately supported by the Internet, which ischaracterized mainly by variations in throughput and packet loss. Thus, studies on videocoding techniques for transmission over the Internet are of both theoretical significanceand practical value. In this dissertation, special emphasis is given on the fastimplementation of video compression and error-resilient video coding from theviewpoint of source coding. The main contributions and innovation points are asfollows:1. A fast algorithm is proposed to reduce the computational load of the mode decisionwithin H.264 encoder. Candidate modes for current macroblock are first inferredfrom coded adjacent macroblocks by adopting motion information and ratios ofdefined mode sub-sets as inference measures. Then a partial rate-distortionoptimization approach is applied to the candidate modes to make target modeselection. For concerned test image sequences, the proposed algorithm can reduceencoder complexity significantly while coding quality is only slightly decreased.2. The Flexible Macroblock Ordering (FMO) feature specified in H.264 provides anopen, flexible framework for arbitrary slices coding, which is intended to intensifyerror resilience. A new FMO implementation scheme is proposed in which theforeground and background slice group pattern for current picture is built on motionanalysis of previous coded picture. Since the foreground is of potential interest, theforeground slice group is further divided into dispersed slices for the sake of higherror resilience. The proposed scheme performs superior to the pattern-fixed sliceallocation types in tradeoff between error resilience and coding efficiency.3. A modified scheme called Perceptual Slice Group based Multiple Description videoCoding (PSG-MDC) is proposed to improve error resiliency of H.264 stream. Bothmain and redundant slice groups are alternate in each description for detailed areaswhile fixed (either main or redundant) slice group is adopted for relativelyhomogeneous areas. By using PSG-MDC, the quality of the reconstructed videocan be improved while the perception is taken into account.4. A method of selective enhancement for Fine Granular Scalability (FGS) is proposed.Our analysis shows that, for image sequences with particular characteristics,Region-of-Interest (ROI) may contain a significant part of areas with high motiondetail. This feature is used in the proposed FGS coding in which bit planes ofmacroblocks with small partition sizes are bit-shifted up prior to encoding in orderto give them a higher priority, hence maximizing the quality of the potential ROI inthe decoded video.5. Two synthetic error-resilient video coding schemes are proposed: â‘ ModifiedMultiple Description Scalable Coding (MDSC) scheme. Multiple DescriptionCoding (MDC) process is applied to the base layer of a scalable coding module,and then each derived description is encoded by another scalable coding module.The advantage of the modified scheme is that both coding efficiency and errorresilience of the coded stream can be considered simultaneously;â‘¡ModifiedMultiple Path Transport (MPT) scheme for FGS video. The bit-planes of theenhancement layer are transferred by a way that is selected adaptively from choicesof duplication and MDC according to bandwidth variation of the paths. Theadvantage of the modified scheme is that both error resilience and bandwidthflexibility of the coded stream are considered simultaneously.