Evaluation Of Prediction Structures For Multiview Video Coding And Optimization Of Stereo Video Encoder

As a novel digital media of vision, multiview video meets the users needs of selecting a certain view from multiple views and operating the audiovisual object, as a result, it provides the 3D depth impression of the observed scenery and the interactive operation function. It would be the next generation of multimedia application after the high-definition flat-panel TV.This paper focuses on the research of the representation of multiview video and the evaluation and optimization of multiple coding functions of multiview video technology, especially the function elvaluation, which including compress feature, random access feature and scalability, and optimization design of the prediction structure of multiview video, and the optimization of stereo video encoder. The main achievements of this paper are listed as follows:Basing on the research of the influence of the prediction structure on multiple coding features, couples of reasonable and efficient quantitative elvaluation model are proposed. For the random access feature, basing on the testing results of user habits of random access, an interactive random access model is proposed. The evaluation model of the relationship between multiview video prediction structure and random access function is formed over it. An evaluation model for scalability is formed after carefully analyzing of different types of scalability, which including time scalability, space scalability quality scalability and view scalability. Finally, a multi-objective comprehensive evaluation model is proposed. Under the guidance of this model, multi-objective optimization on multiview video prediction structure is implemented successfully.Against the application of stereo video coding, a research of optimizing implementation was done basing on the reference software JM17.2. A region of interest (ROI) based adaptive rate-distortion control algorithm is proposed. Firstly, the proposed algorithm introduces the stereo matching method to figure out the ROI. Secondly, the proposed algorithm adjusts the quantitative parameter (QP) of each macroblock. By using this algorithm, the total rate of the coded bit stream can be reduced by 5.05% on average while maintain the objective quality of video.Against the demand of engineering application, an stereo video extend is made basing on the existing x264 encoder, which at the aim of supporting the standard of MVC-3D. As shown by experiment results, the stereo video encoder that basing on the x264 encoder is superior to JM17.2 at the aspect of compression feature and encoding time. The encoding time the former need is just one eight hundredth of the later. For the most of test saquences, real time encoding can be implemented on common computer platform at the condition of relative large QP.A new way of representation basing on the video object (VO) for 3D scenery is proposed. This representation first divides a 3D scenery into several different object layers, then integrates the data in different views which belonging to the same object layer into one object layer of main view. Finally, the corresponding depth information of each object layer is expressed by a depth function or the combination of a depth level and a depth change mode. The proposed representation would be an ascendant one for its high compression efficiency, elimination of occlusion and low complexity of decoding, and is suitable for the coding and application of 3D video.