Adaptive Wavelet Transformation for Image Coding Using a Forecast Decomposition Selection

This paper presents a new approach to an adaptive wavelet transformation. Contrary to the known techniques for wavelet packets, relevant statistical properties (mainly the correlation) of the current subband are rst analyzed. Dependent on that, the decomposition decision is made, whether the subband should be transformed or not. This procedure does not yield to a best-basis selection but to a near-optimal decomposition structure. Furthermore the decomposition structure is generalized. Due to the instationarity of images as two-dimensional signals, the decom-positions of rows and columns are performed irrespective of each other, and it is possible to transform spatial parts of subbands.