Distributed source coding for image and video applications

Many video compression schemes (e.g., the recent H.264/AVC standard) and volumetric image coding algorithms are based on a closed-loop prediction (CLP) framework. While CLP based schemes can achieve state-of-the-art coding efficiency, they are inadequate in addressing some important emerging applications such as wireless video, multiview video, etc., which have new requirements including low complexity encoding, robustness to transmission error, flexible decoding, among others. In this research we investigate new video and image compression algorithms based on distributed source coding (DSC), and we demonstrate the proposed algorithms can overcome some of the deficiencies in CLP based systems while achieving competitive coding performance.;The first part of this thesis discusses our work to explore DSC principles for designing hyperspectral imagery compression algorithms, with an eye toward an efficient and parallel encoder implementation with modest memory requirement. Using DSC tools allows encoding to proceed in "open loop'', and this facilitates parallel compression of spectral bands in multi-processors configurations. We demonstrate that our proposed DSC techniques can be adaptively combined with set partitioning of wavelet coefficients to exploit spatial and spectral correlation. Our latest results show the proposed algorithm can achieve a comparable coding efficiency to a simple 3D wavelet codec developed at NASA-JPL.;The second part of this thesis investigates DSC based coding algorithms to address the flexible decoding problem in video applications. In these, the encoder needs to compress a current frame under uncertainty on the predictor available at decoder. Flexible decoding is relevant in a number of applications including multiview video, frame-by-frame forward and backward video playback, robust video transmission, etc. The proposed algorithm incorporates novel macroblock mode switching and significance coding within the DSC framework. This, combined with a judicious exploitation of correlation statistics, allows us to outperform other competing solutions.;The third part of this thesis proposes solution to address the correlation estimation problem in DSC, which is an important subject for practical DSC applications. We formulate the rate-constrained correlation estimation problem in a DSC framework, and propose information exchange strategies that minimize the rate penalty due to inaccurate estimation. We also propose a novel model-based method for correlation estimation in the context of DSC. We demonstrate that the model-based estimation can achieve accurate estimation with minimal computational and data exchange requirements.