BAMS METHOD: THEORY AND SIMULATIONS

Brani Vidakovic, Duke University
and
Fabrizio Ruggeri, CNR-IAMI

In this paper we address the problem of model-induced wavelet shrinkage. Assuming the {\it independence} model according to which the wavelet coefficients are treated individually, we discuss a level-adaptive Bayesian model in the wavelet domain that has two important properties: (i) it realistically describes empirical properties of signals and images in the wavelet domain, and (ii) it results in simple optimal shrinkage rules to be used in fast wavelet denoising. The proposed denoising paradigm {\bf BAMS} (short for Bayesian Adaptive Multiresolution Smoother) is illustrated on an array of Donoho and Johnstone's standard test functions and is compared to some standard wavelet-based smoothing methods.

The manuscript is available in postscript and pdf formats