Optimal refinement/significance map tradeoffs in SPIHT-based image compression

A rate-distortion-optimal embedded image compression algorithm is one in which each bit is generated by the encoder in such a way that the reconstruction error of the decoded image is reduced as much as possible when that bit is received. Popular embedded algorithms like set partitioning in hierarchical trees (SPIHT) use heuristic techniques to approximately achieve such optimality. Two other approaches have been developed recently that are rigorously optimized for rate-distortion within their respective algorithmic frameworks. In this work, we address the question of optimality within the framework of SPIHT, focusing specifically on the ordering of refinement and significance map information within the bit stream. From our experimental results, we conclude that SPIHT is almost optimal with respect to its ordering of these passes.