Algorithms for scalable and network-adaptive video coding and transmission

Abstract

Real-time multimedia services over the Internet face some fundamental challenges due to time constraints of those applications and network variations in bandwidth, delay and packet loss rate. Our research addresses the problem of network-adaptive video coding and streaming based on source codecs that provide scalability to match the network environments. The first part of the thesis focuses on scheduling algorithm design for network-adaptive video streaming. We extend previous work on rate-distortion optimized video streaming to address more general coding techniques that support multiple decoding paths to enhance adaptation flexibility. Prior work had only considered a single decoding path. Examples of multiple decoding paths include cases where there are multiple redundant representations of the media data or where error concealment is used. An example of these codes is our proposed multiple description layered coding (MDLC), which combines the advantages of layered coding and multiple description coding. This work is composed of several main components: (1) a new source model called directed acyclic hyperGraph (DAHG) to estimate the expected end-to-end distortion; (2) rate-distortion based scheduling algorithms to adjust dynamically the system's real-time redundancy to match the channel behavior; and (3) performance analysis on both source redundancy and transport redundancy. Experimental results show that the proposed streaming framework can provide a very robust and efficient video communication for real-time applications over lossy packet networks. The second part proposes a framework for adaptive scalable video coding using Wyner-Ziv techniques. The current scalable video coding standards suffer to some degree from a combination of lower coding performance and higher coding complexity, as compared to non-scalable coding. A key issue is how to exploit temporal correlation efficiently in scalable coding. We propose a novel scalable coding approach by introducing distributed source coding in enhancement layer prediction in order to achieve a better coding performance with reasonable encoding complexity. Experimental results show significant improvements in coding efficiency over MPEG-4 FGS, especially for video sequences with high temporal correlation.