Multiple description wavelet coding with dual decomposition and cross packetization

Abstract

In this paper, a novel wavelet-based multiple description coding (MDC) is proposed with dual decomposition (DD) and cross packetization (CP). Through dual decomposition, each description has two parts, the primary and the complementary. The former contains the structural information, including the positions and signs of significant coefficients, whereas the latter contains the residual data, which are the absolute values of the significant coefficients’ magnitude. The primary part is most crucial for source reconstruction; hence, their codes generated by the x-tree wavelet encoder are duplicated in both descriptions for heavy protection. On the other hand, the residual data are processed by the multiple description scalar quantizer to generate two indices for their respective descriptions as the complementary part. The proposed DD–MDC method effectively enhances error-resilience capability for robust transmission. For the packet-switching networks, the proposed CP, which produces the row packets and column packets for the two descriptions, respectively, is incorporated into the DD–MDC framework, leading to the DDCP–MDC scheme. Extensive simulation results have clearly shown that the proposed wavelet-based DD–MDC and DDCP–MDC methods are highly error resilient and consistently yield high coding gain; particularly, when only one description is available, the reconstructed image quality is superior to that obtained with other existing methods.