Research On Multi - Description Video Coding Based On

With the development of communication and network technology, therequirements for high quality, real-time video transmissions via network arebooming (for example the wide use of the service on3D video). However, toefficiently transmit such a mass of video data, compression is required. Sincecompressed video stream is very sensitive to transmission errors, such as thepacket loss, bit error, and transmitting delay while transmitting on theunreliable networks，So a good video coding algorithm should possess notonly high compression but also a good error resilient performance. Therefore,the video coding algorithms with high compression and high error resilientability are studied in this dissertation. Based on the analyzing andsynthesizing of the state-of-the-art of the H.264based multiple descriptionvideo coding algorithm, three novel algorithms with high error resilient abilityand coding efficiency are proposed.First, a Mode-Copying Multiple Description Video Coding algorithm isproposed based on the Prediction-Based Spatial Poly-phase TransformMultiple Description Video Coding. The input video sequence is firsthorizontally and vertically down-sampled to form four sub-sequences, andthen, the resulted four sub-sequences are pair-wisely grouped to form twodescriptions. Due to the strong spatial correlation and temporal correlation ofthe two sub-sequences in one description, the best modes and motion vectorsin corresponding macro-blocks are basically the same. Only one sub-sequenceper description is coded by a H.264coder, and the other sub-sequence iscoded by using the best modes and motion vectors of the aforementionedencoded subsequence. The experimental results have shown that the proposedalgorithm achieves a higher coding quality compared with other two H.264based MDC algorithms at moderate and high bit rates.To further eliminate the temporal correlation and spatial correlationbetween the sub-sequences, another new multiple description video codingalgorithm is presented in this paper. Firstly, the given macro-block is encoded by the Prediction-Based Spatial Polyp-phase Transform Multiple DescriptionVideo Coding algorithm and the Mode-Copying Multiple Description VideoCoding algorithm, respectively, and their SADs are separately calculated. Theone with smaller SAD is adaptively selected as the coding scheme of thecurrent macro-block. The empirical results have shown the good performanceof the proposed adaptive multiple description coding compared to the twoalgorithms.Since spatial dislocation exists between two subsequences in the samedescription, the best modes of the macro-blocks with the edges may bedifferent which results in the decline of coding efficiency. To improve theaccuracy of mode-copying algorithm, a novel Multiple Description VideoCoding based on edge correction is also proposed, and further improves theeffect of multiple description coding based on the mode of replication.