Video and image coding for broadband integrated services digital networks

The growing demand for visual telecommunication services over the last decade has greatly increased the need for efficient image coding and compression schemes. The work presented in this thesis examines several aspects of the problem of coding and compressing the various image and video-based services which are likely to utilize Broadband Integrated Services Digital Networks (BISDNs) in the future. This research has two major thrusts. The first is the development of a general-purpose, high-performance and high-quality image coding and compression scheme for these broadband-based visual services. The second is the development of an accurate traffic source modelling scheme for the Variable BitRate (VBR) packet video traffic produced by the proposed coding scheme. The proposed high-performance visual coding and compression scheme combines both statistical and psychovisual coding techniques to produce an optimum scheme which removes both statistical and psychovisual redundancies from images in the coding process. When used to encode four standard 512x512x8-bit test images, this scheme results in sub-distortion compression ratios of up to 27:1 without the use of any form of interframe coding. An efficient and flexible Asynchronous Transfer Mode (ATM) cell packaging scheme is also developed which allows the production of either 'priority' or 'non-priority' cell streams suitable for transmission over broadband networks. Three suitable VBR packet video traffic source models are developed during this study. Each of these sources reproduces the 'low-level' cell generation process and switches between different 'modes' of cell generation in order to capture the inherent `burstiness' and 'variability' of typical VBR packet video traffic streams. However, the way in which this cell production process is modelled, as well as the traffic levels which can be reproduced, differs for each of these sources. A relatively complex, general-purpose traffic source is proposed which is based on the hidden Markov statistical model and is able to model all levels of VBR traffic (up to 20 Mbps). A simpler 'switched-fractal' source is also proposed as an accurate model for the low (below 5 Mbps) VBR traffic only. The third proposed source is a 'switched-Markov' model which specifically caters for the high (5 to 20 Mbps) levels of VBR traffic. Among these three artificial traffic sources, the characteristics of all levels of traffic produced by the proposed video and image coding scheme are able to be reproduced accurately.