Near-best basis selection algorithms with non-additive information cost functions

Search algorithms for finding signal decompositions called near-best bases using decision criteria called non-additive information costs are proposed for selecting bases in wavelet packet transforms. These new methods are compared with the best bases and additive information costs of Coifman and Wickerhauser (see IEEE Trans. Information Theory, vol.38, p.713-18, 1992). All near-best and best bases were also compared with the matching pursuit decomposition of Mallat and Zhang (see IEEE Trans. Signal Processing, vol.41, p.3397-3415, 1993). Preliminary experiments suggest that for the application of time-frequency analysis, a wide variety of results can be obtained with the different methods, and that for the application of data compression, the near-best basis selected with non-additive costs may outperform the best basis selected with additive costs