Design Of Source Decoder And Research Of Multi-view Video Coding

With the development of technologies, digital media has influenced and changed modern society over the last 2 decades significantly. The digital television is a notable representative of the various applications of digital media. A corresponding industrial chain has become large-scale with the popularization of digital television. Just in the transition period of analog video to digital and high-definition formats, new types of media such as 3D video have attracted people's attention by their subtle charms.This dissertation is focused on two issues. First, taking the digital television source decoder as the research subject, it presents the design and implementation approach of the key parts including transport demultiplexor and video decoder. Second, considering the application of 3D and interactive video in the future, this dissertation proposes several efficient algorithms on the performance modeling and optimization in multi-view video coding (MVC).The main work and contributions are as follows:A high-performance design and implementation methodology of twin transport demultiplexor is presented, which is able to demultiplex two transport streams and multiple audio/video elementary streams synchronously. The key parts of the architecture are shared by the two TS inputs in terms of every-cycle time slice and every-section time slice. This design is sufficient to support more tremendous functions with little additional resource.A novel video decoding architecture is proposed, which can support multiple standards including MPEG-2,MPEG-4,H.264/AVC, AVS and RealVideo 8/9/10.An elaborate software and hardware cooperation scheme is adoted for the tradeoff of flexibility and real-time performance. High speed achitectures are designed for the most computationally expensive parts in video decoder, including AVS context-based binary arithmetic decoder and motion compensation sub-system for multiple standards.On top of this, research on MVC is also explored in this dissertation. The proposed directed tree decomposition offers a new perspective to analyze and arrange hierarchical B prediction structures. Besides, a linear model based on reference intervals is proposed for the quantitative evaluation of B pictures and hierarchical B prediction structures. To evaluate the relative efficiency of prediction structures, the model can be further simplified by omitting parameters. In combination with the proposed linear model for compression efficiency, it is straightforward to evaluate the performance of any hierarchical B prediction structure from the directed tree.With a dynamic programming method, the optimal tree is set up elegantly, which can be used for tradeoff optimization of compression efficiency and random access ability. Experiment results show that the optimal tree of compression efficiency achieves higher performance than the existing hierarchical B prediction structure.Furthermore, this dissertation proposes a novel method to optimize the inter-view prediction structures for MVC.The construction of prediction structure is converted to the arrangement of coding order. Then simulated annealing (SA) algorithm is employed to minimize the total cost in order to obtain the best coding order. Experiment results show that the annealing process converges rapidly to satisfactory states in this problem. SA algorithm exploits the inter-view correlation to the best of its potential and the generated optimal prediction structures achieve 0.1-0.8 dB higher PSNR performance than the reference prediction structure of MVC.The proposed method is applicable to arbitrary camera arrangements.