Ultrasound speckle suppression using heavy-tailed distributions in the dual-tree complex wavelet domain

Abstract

A complex wavelet-based Bayesian method is proposed for denoising of medical ultrasound images. The symmetric alpha-stable distribution (SaS) is used to model the real and imaginary parts of the complex wavelet coefficients of logarithmically transformed noise-free images. The coefficients that correspond to the noise are assumed to approximate a Gaussian distribution. These models are then exploited to develop a Bayesian maximum a posteriori (MAP) estimator, which is well defined for all SaS random variables. To estimate the wavelet coefficients statistics precisely and adaptively, we classify the wavelet coefficients into different clusters using context modeling, which exploits the intrascale dependency of wavelet coefficients. The simulations demonstrate an improved denoising performance over some related earlier techniques.