On the choice of 'wavelet' filter for still image compression

The author addresses how regularity, frequency selectivity, and phase act upon the performance of a still image compression scheme using wavelet decomposition. Preliminary results are given for a simple compression scheme using orthonormal separable wavelet transforms, scalar quantization, rate/distortion optimization, various coding criteria, and a large number of 'wavelet' filters with balanced regularity, frequency selectivity, and phase. It is shown that regularity may be relevant for still image compression, at least for short filters (L<or=12), for which the regularity order is relatively small. Using more regular filters is probably useless, as the compression performance does not greatly improve for longer filters. Moreover, the effect of phase seems negligible for orthonormal filters.<<ETX>>