Multiple Description Coding with Edge-Adaptive Reconstruction

In this paper we present a framework for estimating corrupted coefficients in multiple description coding (MDC) of images. Descriptions are generated by partitioning the domain of wavelet coefficients into sets whose elements are maximally separated from each other. Consequently an edge-adaptive estimation algorithm is proposed to explore the local texture statistical characteristics for estimating the lost coefficients. The edge adaptation property of our algorithm significantly increases the intra-subband dependencies employed in reconstruction. Extensive experimental results show that impressive improvements on both objective and subjective measures can be achieved. Key-Words: Multiple description coding, error concealment, covariance method, domain partitioning based MDC, edge-adaptive algorithm, mutual information.