Bayesian denoising of ultrasound images using heavy-tailed Levy distribution

Abstract

Ultrasound images often exhibit poor signal to noise ratio when compared with optical images, because of the presence multiplicative speckle noise. Speckle suppression is often carried out as a pre-processing step to aid in diagnosis using ultrasound images. In this study, dual tree complex wavelet transform-based Levy Shrink algorithm is proposed for denoising ultrasound images. The coefficients in each wavelet subband are modelled using a heavy tailed Levy distribution. The scale parameters of the Levy distribution are estimated using fractional moments. Within this framework, a Bayesian estimator is employed to denoise ultrasound images. The proposed Levy Shrink algorithm is verified using evaluation parameters such as peak signal to noise ratio, mean structural similarity index, correlation coefficient and equivalent number of looks. The efficiency of the proposed denoising algorithm is justified by conducting extensive experiments on real as well as simulated ultrasound images.