3-D multiple description video coding for packet switched networks

Abstract

Multiple description coding (MDC) schemes have become popular as robustness tools in packet switched networks where multiple paths with varying packet loss rates are available. In this paper we propose a new MDC scheme for video using 3D wavelet decomposition. The spatio-temporal coefficients are then grouped using either a random, uniform scheme or a hierarchical scheme. The descriptions are constructed by arranging a subset of these groups in two positions: primary and the secondary. The bit-allocation problem between the primary and secondary groups is formulated and solved using the generalized BFOS [E.A. Riskin, 1991] algorithm. The bits allocated to the secondary position may be zero, in which case, the description is said to be constructed with no redundancy. Experiments using bi-orthogonal wavelet for the spatial and Haar wavelet for the temporal decomposition, along with embedded block coding with optimal truncation for the different groups have shown significant performance for test video sequences. We expect that, for finite packet loss probability, the MDC scheme with redundancy does better than the scheme that does not have redundancy. In this paper, we also present a discussion of the differences between our work with the most recent 3D based video MDC scheme [Joohie Kim et al., 2002].