A low memory wavelet zerotree coding for images transmission over noisy channels

This paper presents a Robust Low Memory Zerotree Coder (RLMZC) with much less working memory. Two distinct bit sequences are designed according to their importance and channel noise sensitivity levels then being protected differently. The RLMZC algorithm abandons the use of lists in most of zerotree coding methods, defines a compact flag map, and introduces the heuristic depth finding strategy in the significant location operation and significant refinement operation. The compact flag map stores the status of coefficients and the heuristic depth finding strategy searches the significant descendant coefficients in tree branches. Furthermore, the proposed algorithm also provides a coding framework that separates the significance bits from the refinement bits, which helps to employ unequal error protection over noisy channels. Comparison of RLMZC with Set Partitioning In Hierarchical Trees(SPIHT) shows that RLMZC saves at least 1.085Mbytes of memory but only reduces minor peak signal-to-noise ratio(PSNR) values for coding a 512 ×y image. For noisy channels our coder is more robust than SPIHT with equal error protection. In addition, RLMZC outperforms another Low Memory Coder (LZC) in both noisy and noiseless channels. The RLMZC is shown to be highly promising for some memory limited applications and noisy channels transmission applications.