| In this dissertation two application layer quality of service control mechanisms, rate control and error control , are addressed in order to deal with the requirements associated with real time video communications applications and effective yet computationally simple solutions are brought forth under the "implementable in real-time" constraint.; Specifically, first, the problem of rate adaptation of pre-compressed video bitstreams to meet the bandwidth constraints, which are either dynamically varying or only known at the time of transmission, is addressed. Effecting rate adjustment on pre-compressed video by operating directly on the compressed domain is commonly referred to as rate shaping, and in this work it is realized through the signal-to-noise ratio (SNR) scaling of an already encoded video object to an alternate (lower) rate-(higher) distortion level. For the SNR scaling of transform data, a family of practical algorithms is devised, based on the rate-distortion optimal selection of the transform coefficients, which perform very close to the best result achievable by an exhaustive, and therefore unfeasible in real-time, search of the solution space. For the SNR scaling of motion data, a new scalable representation for the motion vector information and a novel algorithm for its efficient coding are developed. These scaling algorithms provide progressively varying quality within an acceptable range on a continuum of bit rates.; As the second problem, the concealment of lost transform data is considered. A family of intra-frame error concealment algorithms is developed. These algorithms accurately reconstruct the visually significant local image features by exploiting the directional nature of the inter-neighboring-block coefficient correlations in identical transform channels as well as the even/odd symmetry properties of discrete cosine transform's basis functions. Estimating an optimal DC value satisfying a maximal pointwise smoothness criterion and then applying a deblocking filter further enhance the reconstruction quality, leading to visually very pleasing results.; Operating directly in the compressed domain, all of the algorithms have extremely low complexity and highly amenable to both hardware and software implementations. |
