Recovery of Lost VQ Indexes in Packet Transmission

We consider the problem of robust transmission of VQ-coded image/video via noisy packet networks. In the event of packet loss, some VQ index bits will be absent at the receiver side. But the very knowledge of lost packets identifies the spatial locations of affected VQ blocks, which are powerful information for the decoder to estimate the missing VQ index bits. This is possible because of the statistical redundancy between spatially adjacent VQ blocks. In this paper we present a MAP (maximum a posterior) estimation technique to recover missing VQ index bits due to packet loss. The novelty of this work is to couple high-order Markov modeling with MAP while avoiding the problem of context dilution.