Distributed compression of zerotrees of wavelet coefficients

A distributed coding algorithm is presented for compression of wavelet-transformed data. Data structures based on zerotrees are exploited for efficient compression of the significance map of wavelet coefficients. The coefficients are scanned in two stages, with a significance pass and refinement pass, similar to the SPIHT algorithm. The bits resulting from these passes are Slepian-Wolf coded using an LDPC syndrome code selected from a bank of available codes. A key realization is that, for each bitplane of the wavelet coefficients, the significance pass of the source data can be synchronized with that of the side information. This allows distributed compression of the significance pass. This is substantially different from previous mixed approaches in which the refinement pass was Slepian-Wolf coded, but the significance pass was coded independently. Rate-distortion results are presented for images from the ALOS AVNIR-2 multispectral dataset and compared against those obtained with SPIHT and JPEG2000.