Video compression and streaming over packet-switched networks

This dissertation deals with video streaming over (a) best-effort unicast and multicast packet networks, and (b) packet networks with traffic differentiation. First, we consider the problem of developing a scalable video compression scheme that is resilient to random packet loss. By using three-dimensional subband decomposition and data partitioning in the subband coefficient domain, we demonstrate a packetization scheme in which each packet contains information about disjoint partitions of subband coefficients. In such a way, individually decodable packets of equal visual importance can be obtained. In actual experiments over the Internet, our scalable compression technique is found to achieve significantly lower average distortion and smaller variation in image quality compared to schemes using non-resilient scalable compression.; Second, we explore the transport protocols necessary for transporting scalable video over best-effort unicast and multicast packet networks. For unicast, we construct a TCP-friendly transport protocol that shares bandwidth fairly with multiple instances of itself and TCP. Our TCP-friendly protocol, however, differs from TCP in being low-latency and without abrupt changes in communication rate. For multicast, we propose the layered Forward Error Correction (FEC) scheme that allows different receivers in a multicast environment to efficiently receive different levels of protection commensurate with their respective channel conditions. For rate control at the receivers, we propose an equation-based approach that computes network usage as a function of measured network characteristics. We show that equation-based rate control achieves more fair bandwidth sharing as compared to existing multicast rate control schemes. Simulation and MBONE experiments have shown that video quality is significantly improved when layered FEC is used.; Third, we address the problem of transmitting compressed MPEG video over packet networks with traffic differentiation in terms of loss and delay. To this end, we present a framework in which MPEG compressed video is divided into sub-streams of different delay and loss sensitivities. These substreams are then transported using different quality of service to improve overall network utilization. We present a number of packet classification schemes for MPEG bit-stream based on delay and loss characteristics of the data, and compare those schemes using commercial DVD contents. It is shown that significant reduction in distortion or delay can be achieved compared to schemes that treat the MPEG stream as homogeneous in importance.