Robust and multiresolution video delivery : From H.26x to Matching pursuit based technologies

With the joint development of networking and digital coding technologies multimedia and more particularly video services are clearly becoming one of the major consumers of the new information networks. The rapid growth of the Internet and computer industry however results in a very heterogeneous infrastructure commonly overloaded. Video service providers have nevertheless to oer to their clients the best possible quality according to their respective capabilities and communication channel status. The Quality of Service is not only inuenced by the compression artifacts, but also by unavoidable packet losses. Hence, the packet video stream has clearly to fulll possibly contradictory requirements, that are coding eciency and robustness to data loss. The rst contribution of this thesis is the complete modeling of the video Quality of Service (QoS) in standard and more particularly MPEG-2 applications. The performance of Forward Error Control (FEC) algorithms is rst analyzed in terms of the Packet Loss Ratio and Average Burst Length of video data after error recovery. This study leads to a mapping of the network packet loss into video degradation. Along with the analysis of the source rate-distortion characteristic of the video coder, it results in a set of equations that completely model the video quality of service as a function of the scene content and the transmission conditions. An optimal rate distribution between source and channel coding is nally issued from this complete QoS model. Even with an optimal channel rate, remaining losses may still generate severe damage to the decoded video. All current video standards are based on motion estimation techniques and dierential coding which make them really vulnerable to data loss. To avoid error propagation within the displayed sequence, a new adaptive structuring and protection algorithm is applied to MPEG-2 video coding. This scheme basically triggers the insertion of syntactical redundancy (i.e., Intra-coded macroblocks and slice headers) to stop the spreading of data loss within the hierarchical video stream. Redundancy is thus inserted where most needed trading-o compression efficiency against error resilience. The most important video packets are additionally protected by FEC packets to avoid the loss of critical data. This complete error resilient coding scheme is shown to outperform other standard-based robust algorithms. Although video standards are necessary to the deployment of video applications, they do not provide sucient robustness to unavoidable packet losses. Moreover, they truly lack a charac- teristic considered as mandatory in nowadays heterogeneous framework, that is an ecient and ne granularity multi-resolution coding. To cope with these limitations, a new way of coding video based on Matching Pursuit is proposed in the second part of this work. The sequence is decomposed into a series of spatio-temporal functions from a redundant structured dictionary. A new formulation of the structural redundancy is proposed to evaluate the performance of non- orthogonal bases in signal decompositions. Several new spatio-temporal dictionaries are then proposed and compared in terms of compression eciency and decoding complexity which is an important issue for heterogeneous decoding devices. A tentative comparison with wavelets and video standard coding shows that Matching Pursuit coding is at least as efficient in terms of compression eciency and even more in low rates. Additionally Matching Pursuit oers intrinsic robustness and multi-resolution properties. The properties of Matching Pursuit streams are then used in the design of a specic a posteriori quantization scheme. This novel algorithm only depends on the dictionary characteristics and offers signicant improvements over classical coecient quantization methods. It is shown also that index quantization should be preferably avoided, hence emphasizing the need for a careful design of the dictionary as a compromise between compression eciency and coding rate. Finally, the lossy and heterogeneous Matching Pursuit streams delivery is studied from the viewpoint of transcoding and error control. A new transcoding scheme is proposed as a simple algorithm to adapt structured Matching Pursuit streams to the resolution of the receivers. An optimal Unequal Error Protection scheme is lastly proposed to ght against packet losses. The end-to-end error control may however become relatively complex in very disparate environments, so that a novel Equivalent Error Packetization is designed to ease the video distribution management. All these dierent building blocks show that video Matching Pursuit coding constitutes a very promising alternative to classical schemes in lossy and heterogeneous environments.