Politecnico Federale Di Losanna Swiss Federal Institute of Technology Lausanne

This work addresses the quality of service (QoS) analysis and optimization of TV-resolution MPEG-2 video streams to be transmitted over packet networks facing a given packet loss probability. The first contribution of this work resides in the analysis of the end-to-end QoS of MPEG-2 video streams transmitted over a lossy packet network. We address the problem of video quality prediction and control for high resolution packet video. We analyze how the user-perceived quality is related to the average encoding bitrate for VBR MPEG-2 video. We also show why simple quality metrics (e.g., PSNR) may lead to unconsistent interpretations. Furthermore, for a given coder setup, we analyze the effect of packet loss on the video quality. We then propose a parametrization of the QoS as perceived by a customer. We finally demonstrate that, when jointly studying the impact of coding bit rate and packet loss, the reachable quality is upper-bounded and exhibits one optimal coding rate for a given packet loss ratio. We show that this quality-optimal encoding rate may be tuned via both source and channel coding and/or protection. Therefore the second part of this work tackles the optimization of the end-to-end QoS by increasing the video stream robustness to data loss while conforming to the MPEG-2 syntax. We propose a new algorithm which adaptively inserts spatial and temporal resynchronization points (i.e. slice headers and intra-coded macroblocks respectively): the Adaptive MPEG-2 Information Structuring (AMIS) algorithm. In general, intra-coded macroblocks require higher bit rates compared to their non-intra coded versions. Similarly, slice header insertion adds overhead. There is obviously a tradeoff between the encoding quality and the robustness of an MPEG-2 video stream. We compare the performance of AMIS against a classical MPEG-2 encoding scheme (i.e., the Test Model v5). All the quality assessments are based on a perceptual quality metric, MPQM 1, which proved to behave consistently with human judgements. Results show that the proposed mechanism behaves much better in terms of global perceptual video quality.