An algorithm for adaptive selection of the best frequency, or wavelet packet basis was proposed by Coifman, Quake, Meyer and Wickerhauser [CMQW90]. The wavelet packet algorithm generates a library of orthonormal functions that are derived from a single filter kernel. The wavelet packet algorithm searches through the library to find the least cost basis which also provides the best compression. By using a filter bank the wavelet packet library is produced by cascading filtering and downsampling operations in a tree-structure. The tree also guides the search for best basis [CMQW90][Wic90][RV93]. But an important drawback of wavelet packets is that the decomposition is performed on the entire image or on fixed blocks of the image. Wavelet packets do not adapt to variations in content across the separate regions of the image or to non-stationarity.