可調性匹配搜尋視訊壓縮演算法與階層式群播技術

Abstract

It has been shown that matching pursuit video coding achieves a better performance than DCT-based video coding in terms of subjective and objective quality at very low bit-rates. However, due to its massive computational complexity, traditional matching pursuit video encoding is usually only approximated by searching a local maximum atom at each iteration. In this thesis, we introduce a multiple blocks update algorithm for matching pursuit video coding that achieves better coding performance and lower coding complexity than traditional approaches. We also introduce a systematic dictionary design, which combines eigen-dictionary approximation and tree-based atom search. This systematic approach approximates a given target dictionary efficiently and enables a tradeoff between coding performance and complexity. A numerical solution is also developed to achieve the optimal coding performance under the constraint of a given complexity. We propose a fine-grained scalable (FGS) matching pursuit video coding scheme based on progressive atom coding and quadtree representation of atom positions. To this end, we develop a quadtree prediction algorithm to explore the temporal and spatial redundancy of atom positions between adjacent bit-planes. We also propose a two-layer matching pursuit video coding scheme in which the same set of motion vectors, estimated from a weighted combination of previous base and enhancement layers, is applied to motion compensation of both layers. The two-layer system achieves a tradeoff between the coding quality of the base and enhancement layers, while retaining their fixed bit-rates. Finally, we present a new, fine-grained scalable layered multicast scheme based on a hybrid coarse/fine layered structure. We develop a congestion control protocol for hybrid coarse/fine layered multicast by using hierarchical layered probing and one-way delay trend detection. We demonstrate that fine-grained rate adjustment can be achieved through hierarchical probing. By combining the fine-grained layered multicast approach with scalable video coding, we present a framework that achieves efficient scalable video streaming over heterogeneous networks.