Research On Real-Time Scalable Video Coding And Rewriting Technology

In realistic application, the processing speed of video players and display capability are of enormous disparity, along with the tremendous difference of network conditions. The emergence of Scalable Video Coding (SVC) aimed at encoding once to generate a bit-stream containing different resolutions, frame rates, and video quality; therefore, users can merely obtain part bit-streams they need for different devices and network situations. However, the complexity of SVC makes it unpractical to be used in real-time application scenarios; further, motion estimation has not fully deployd the relation among neigbouring blocks. This thesis is focusing on scalable video coding techniques, from perspectives of encoding speed and performance, to design real-time optimizing algorithms of encoder and to refine video quality in offline video applications. The main contents of this thesis is as follows:1. Tackling with real-time application scenarios, this thesis presents several parallel algorithm schemes for SVC, including GOP (Group of Pictures) level, temporal layer level, and slice level, to make full use of the computational capability of multi-core computers; therefore, it can achieve satisfactory coding speed and efficiency.2. This thesis also proposes a real-time SVC to AVC rewriting technique. It works on some cases that many devices cannot decode SVC bit-streams whereas only AVC bit-streams are decodable. Therefore, it breaks through the conventional constraint that the SVC to AVC rewriter can only rewrite the bit-streams serially, raising the degree of parallelism while maintaining a good video quality.3. A two-pass encoding method is presented which utilizes the relativity of motion information of neighboring blocks to improve the coding efficiency. It mainly uses simulated annealing to randomly modify motion information of some blocks after all blocks in an entire frame have decided the best modes, enhancing the encoding performance of the whole frame.Experimental results demonstrate that, the proposed algorithm can significantly boost the speed of encoding and rewriting, and the proposed global motion information optimization approach can improve the coding efficiency in the offline cases.