Noise reduction by multiplicative waveforms decomposition

This paper introduces a novel multiplicative noise reduction method based on a particular association of structural and statistical analysis. The structural analysis is performed by a new, multiplicative matching pursuit decomposition (MMPD), that decomposes images containing the intrinsic variation into a nonlinear expansion of waveforms selected from a dictionary of functions. This selection is made in such a way to match best the image local structures. Local statistics evaluation is associated to the MMPD convergence scheme ; then an adaptive algorithm framework for multiplicative noise is deduced. An application to speckle reduction in synthetic aperture radar (SAR) images is described.