On the Performance of Non-Gaussian Distributions in Modelling the Wavelet Coefficients of Medical Ultrasound Images

A major problem concerning ultrasound images is their inherent corruption with speckle noise. Homomorphic wavelet-based methods using parametric models are widely used for despeckling ultrasound images. However, the efficiency of these methods greatly depends on the accuracy of the prior distribution used for modelling of the non-Gaussian statistics of the wavelet coefficients of the log-transformed reflectivity. An extensive study is carried out on the performance of the generalized Gaussian, symmetric alpha-stable and symmetric normal inverse Gaussian distributions in modelling the wavelet coefficients. It is shown that the symmetric normal inverse Gaussian distribution is a more suitable prior than the other distributions.