An iterative speckle filtering algorithm for ultrasound images based on bayesian nonlocal means filter model

Abstract In this paper, we study the problem of suppressing speckle noise in ultrasound images for better clinical diagnosis and subsequent image processes. In order to employ Bayesian nonlocal means filter (BNLMF) model in the circumstance of speckle noise to realize image restoration, we deduce the key probability density function with the help of speckle noise statistical characteristic and then present an iterative filtering algorithm. The first iteration with the noisy image itself being the input of the filtering model generates an initial estimator of the clean image which then further offers a better input of the filtering model. The constantly updated neighbor patches and probability density functions make the filtering result closer to the potential clean one. The healthy iteration process exports the favorable restored image surpassing the results obtained by some typical despeckling methods. Besides, benefit from the blockwise filtering style, pre-patch-selection operation and a small iteration number, the whole algorithm won't consume much time. Various experiments designed for processing the simulated noisy images and the real ultrasound images prove the superiority of the proposed method.