A nonexpansive pyramidal decomposition is proposed for
low-complexity image coding. The image is decomposed through a nonlinear
filter bank into low- and highpass signals and the recursion of the
filterbank over the lowpass signal generates a pyramid resembling that
of the octave wavelet transform. The structure itself guarantees perfect
reconstruction and we have chosen nonlinear filters for performance reasons.
The transformed samples are grouped into square blocks and used
to replace the discrete cosine transform (DCT) in the Joint Photographic
Expert Group (JPEG) coder. The proposed coder has some advantages
over the DCT-based JPEG: computation is greatly reduced, image edges
are better encoded, blocking is eliminated, and it allows lossless coding.