Lossless Subband Image Compression Using Integer Coeecient Filter Banks

A new class of two-channel lter banks having integer lter coeecients is proposed. The integer lter coeecients are obtained by maximizing for subband coding gain. The number of bits required to represent the subband samples are kept low through eecient nonlinear implementation techniques. Several integer coeecient lter banks are tested for lossless image compression. An octave-band (dyadic) frequency partitioning with 6-stages is preferred. In this way, we achieve a progressive image transmission based on multiresolution image representation in a lossless mode. Two diierent coding strategies are investigated. Firstly, enumeration techniques from pyramid vector quantization (PVQ) scheme are adopted. Secondly, a classwise entropy coding scheme is employed. Both coding schemes tend to perform fairly well, however, the classwise entropy coding scheme performs better than the PVQ based scheme. The compression ratios are encouraging compared to the best results reported so far in the literature.