Improved motion-compensated temporal filtering (MCTF) for scalable video coding

Scalable video coding (SVC) is currently being developed as the new MPEG standardization effort of H.264/AVC scalable extension, a.k.a., Joint Scalable Video Model (JSVM). One key element of such scalable video codec is the hierarchical prediction structure for temporal scalability realized either by the lifting-based motion-compensated temporal filtering (MCTF), or by the hierarchical B pictures alternatively.; In this dissertation, we focus on three important goals for scalabilities realized by MCTF. The first goal is to thoroughly review the recent development of MCTF framework, which includes the inherent drift problem faced by conventional hybrid video coder to support scalability efficiently, the importance of motion compensation in the new light of MCTF framework, and the challenges of efficient MCTF implementation.; The second goal is to find the optimized update steps to improve the SVC coding efficiency. To this end, we propose a class of new MCTF update step schemes, which improves the JSVM update steps in two ways: first, generating the accurate update step motion information via motion vector (MV) correlation analysis; secondly, incorporating chrominance information of video sequence to adaptively perform update steps. Furthermore, we enhance the proposed methods by an efficient content adaptive analysis.; Our third focus concerns the effect of update steps on the MCTF coding efficiency. In this regard, we present a theoretical model and study of the MCTF coding efficiency measured in terms of peak-signal-to-noise ratio (PSNR) of the reconstructed sequence. We investigate the effect of update steps on MCTF coding efficiency when skipping update step at both encoder and decoder sides; when performing update step at the encoder side only; and the contribution of non-linear adaptive update control to the overall reconstruction error. By our theoretical model in this dissertation, we show that the MCTF update step at the decoder side does not contribute significantly to the coding efficiency, except at high spatial-temporal resolutions with high bit rates. Consequently, our study provides a theoretical justification for not including MCTF update steps at the decoder side for the current SVC development.