Multiple description coding with prediction compensation

A new multiple description coding paradigm is proposed in this dissertation, by combining the time domain lapped transform, block level source splitting, inter-description prediction, and coding of the prediction residual. The joint design of all system components is developed, and the asymptotic performance in the DPCM case for first-order Gauss-Markov sources is analyzed. Image coding results show that the proposed method can significantly outperform other methods in the literature. The effects of different transforms and the size of the prediction filter are also studied. A number of generalizations of the proposed scheme is studied as well. First, the two dimensional Wiener filter is introduced into this scheme. These filters can improve the performance when the redundancy is low. Secondly, the proposed scheme is extended to three-description and four-description cases. The optimal design and image coding results are reported.