Research On Fast Algorithm And Quality Optimization Method For Scalable Video Coding

With the fast development of network and video coding technology, people are eager to obtain video information via network. To simultaneously deal with the heterogeneous network environment, different end-users, and personal preferences during the video service becomes a real challenge. As an effective solution for this problem, the Scalable Video Coding (SVC) has attracted much attention:SVC research gains attraction due to the facts that it outputs an embedded stream which is scalable in frame rate, spatial resolution, and quality. And, both the video server and the receiver can adaptively extract the target streams according to the application condition, such as network bandwidth, screen size, and processing capability of end-equipment. However, comparing with the traditionally non-scalable video coding techniques, the computational complexity of SVC is huge and increases with the number of coding layer. In addition, with the adoption of inter-layer prediction tool, each layer's individual complexity is further aggravated. Thus, researches on a fast algorithm of SVC which takes into account of all the new features, is important for practical use of SVC. Furthermore, since the scalable bitstream is only partially received and decoded at SVC decoder side, the video quality at decoder side is different from the encoder side. As human being is the ultimate receiver of the video information which is obtained by the decoding process, the video quality at the decoder side should get much more attention and further optimization. The encoder process plays an important role in the quality control, and the bitstream extraction algorithm has an strong impact as well, thus researches on how to improve the quality both in the encoding process and the stream extraction process also become significant. As a result, it is of theoretical significance and practical value to conduct an in-depth research on the fast algorithm and scalable quality improvement of SVC.In chapter 1, the significance of the research work is presented, and the current research status is briefly summarized. Then, the main research contents and the chapter structure of the thesis are introduced. In chapter 2, the key technologies in SVC are studied, and it serves as indispensable background information for the following chapters.In chapter 3, a fast inter-layer prediction algorithm based on the rate distortion cost (RDC) correlation and motion correlation is proposed, which includes fast inter-layer motion prediction (ILMP) and fast inter-layer residual prediction (ILRP). Firstly, both the BLSKIP mode and Direct mode's rate distortion cost (RDC) is taken as a measurement of macroblock level's motion correlation. Secondly, the motion search process for ILMP off condition is being executed, and the output motion vector is used to calculate the distance between it and the prediction motion vector in both ILMP on and off conditions. This distance is taken as a measurement of block level's motion correlation, and the ILMP choice is made based on both macroblock level's and block level's motion correlation. Lastly, both the motion search and motion compensation with the ILRP being operative is conducted in advance. Then, the residual data is used to estimate the RDC for ILRP being on and off conditions. Fast ILRP choice is made based on the ratio of the estimated RDCs. The advantage of the proposed algorithm is that it can be flexibly combined with other fast algorithm and reduce the complexity further.In chapter 4, a fast mode decision algorithm based on early termination is proposed. Firstly, the frequently occurred modes with little computation requirement are examined in advance, which includes SKIP mode and BLSKIP mode. The benchmark modes are predicted from the base layer's mode. If both the local-optimal condition and the monotony of RDC were satisfied, the mode search process would be terminated. Otherwise, the candidate modes are selected based on the inter-layer mode correlation, and later are adjusted according to the result of the previous step. Finally, the refinement search is conducted and the optimal macroblock mode is obtained.In chapter 5, since all current encoding schemes are solely depending on the bitstream extraction algorithm to improve the video quality, a quality optimization method for SVC is proposed, which includes partially decoding based encoder quality optimization and hybrid rate-distortion based bitstream extraction algorithm. First the enhancement layer's residual transform coefficients are intentionally truncated in the encoding process and the remaining coefficients are reconstructed through the entropy encoding, then the partially decoding based RDC is calculated. Both in the base layer's mode decision process and the enhancement layer's mode decision process, the partially decoding based RDC is employed to select the best mode. Moreover, in the bitstream extraction process, each NAL unit's rate-distortion (R-D) is calculated, and the R-D from the current quality position to the highest quality position is accumulated for the same access unit which is taken as the current NAL's hybrid R-D. Finally, the hybrid R-D slope is employed to assign the priority.The final chapter concludes the new achievements of the whole research, and presents the prospect of the future research.