ADAPTIVE TRANSFORMS FOR IMAGE-CODING USING SPATIALLY VARYING WAVELET PACKETS

We introduce a novel, adaptive image representation using spatially va rying wavelet packets (WP's). Our adaptive representation uses the fas t double-free algorithm introduced in [1] to optimize an operational r ate-distortion (R-D) cost function, as is appropriate for the lossy im age compression framework. This involves jointly determining which fil ter bank tree (WP frequency decomposition) to use, and when to change the fitter bank tree (spatial segmentation). For optimality, the spati al and frequency segmentations must be done jointly, not sequentially. Due to computational complexity constraints, we consider quadtree spa tial segmentations and binary WP frequency decompositions (correspondi ng to two-channel filter banks) for application to image coding. We pr esent results verifying the usefulness and versatility of this adaptiv e representation for image coding using both a first-order entropy rat e-measure-based coder as well as a powerful space-frequency quantizati on-based (SFQ-based) wavelet coder introduced in [11].