A heuristic algorithm to find the best tree-structured image decomposition

In the context of uniform filter bank, width of the hand depends upon the order of the subband decomposition, however for non-uniform filter banks there exist infinite possibilities even when the order is fixed. These infinite possibilities can be brought down to a certain number by using tree structured non-uniform decomposition of the spectrum, with a given pair of low pass and high pass filter. We address the problem of finding the decomposition that gives maximum coding gain. This question is extended to two dimensional signal like image. In this work we suggest a heuristic algorithm that avoids the exhaustive search of max coding gain. We consider separable filters. With the assumption of unit norm of each filter in filter pair.