abstract
-
This thesis addresses the rate-distortion optimal packetization (RDOP) of embedded bitstreams into independent source packets for the purpose of limiting error propagation in transmission over packet noisy channels. The embedded stream is assumed to be an interleaving of $K$ independently decodable basic streams. The goal is to partition these basic streams into $N (N
The RDOP problem previously formulated by Wu \emph{el al.} focused on finding the partition that minimizes the distortion when all packets are decoded. The authors proposed a dynamic programming algorithm which worked under both high bit rate and low bit rate scenarios. In this thesis, we extend the problem formulation to finding the partition which minimizes the expected distortion at the receiver for a wide range of transmission scenarios including unequal/equal error/erasure protection and multiple description codes. Then we show that the dynamic programming algorithm of \citep{DBLP:journals/tmm/WuCX01} can be extended to solve the new RDOP problem.
Furthermore, we propose a faster algorithm to find the globally optimal solution based on the divide-and-conquer technique, under the assumption that all \emph{basic} streams have convex rate-distortion curves. The proposed algorithm reduces the running time from $O(K^{2}LN)$ achieved by the dynamic programming solution to $O(NKL\log K)$. Experiments performed on SPIHT coded images further validate that the speed up is significant in practice.