With the increasing popularity of digital video communications over packet switching networks such as the Internet and wireless networks, error control techniques have become a critical research area. Due to its excellent scalability and performance, embedded 3-D zerotree wavelet video compression has great potential in non-conversational scalable video streaming applications. Although error concealment has been widely used as a “last-step” approach to achieve a graceful quality degradation when packet loss is unavoidable, quantitative analysis of its effect and studies on optimization in regard to distortion are far from adequate. Our research objective of is to improve the QoS of the 3-D zerotree wavelet video streaming system with the error concealment method employed at the receiver. Based on 3-D zerotree wavelet video characteristics and the corresponding error concealment method, this dissertation consists of two major topics: optimal packet interleaving under burst lossy networks and joint optimization of retransmission and error concealment.
To combat the significant quality degradation caused by burst packet losses, we propose a near-optimal temporal packet interleaving algorithm that maximizes the received video quality. The proposed algorithm consists of two steps. First, spatial tree-based interleaving is conducted during the independently decodable packetization to disperse the damage of packet loss. Secondly, optimal temporal packet interleaving is applied during the transmission to minimize the expected distortion of the received video. An iterative exchange-based searching algorithm has been developed to find a sub-optimal solution with significantly reduced computational complexity of O(N2). In addition, to allow real-time implementation and further avoid the characteristics estimation of time-varying lossy networks, a novel measurement, “temporal neighbor packet distance,” is proposed as an alternative optimization criteria of the expected distortion.
Inspired by the concept of “content-based retransmission,” the joint optimization of two error control methods, retransmission and error concealment, has been studied. A new content index, namely “reconstruction distortion,” is defined to quantify the packet's importance in retransmission when error concealment method exists in the video streaming system. Based on the defined content index, the content-based retransmission approach is revised to select the best packet-sending scheme that maximizes the received video quality under the given error concealment method.