Redução de ruído speckle em imagens de ultrassom com filtragem não-local e distâncias estocásticas

Abstract

Ultrasound imaging is one of the most used modalities in medicine since it is low cost and has no ionizing radiation. However, ultrasound images are corrupted by a signal-dependent noise, known as speckle, causing a granular pattern that makes it difficult for visual or automatic image interpretation. Many filters have been proposed in the literature for speckle removal, and more recently, a great deal of attention has been focused on non-local techniques, as the non-local means filter (NLM) and the block-matching collaborative filtering (BM3D). The central idea of these recent methods is a measure of similarity between patches, originally proposed as the Euclidean distance for filtering additive white Gaussian noise. In this work, an approach is proposed for reducing speckle noise based on the non-local techniques and using the support of the information theory. Stochastic distances are derived for Rayleigh, Nakagami and Fisher-Tippett distributions, and used as similarity measures in NLM and BM3D filters. Two methods are applied to generate these stochastic distances. The first method is based on symmetrized statistical divergences and the second one related to the geodesic distances in probabilistic spaces. Although similar approaches already exist to despeckle synthetic aperture radar (SAR) images, here they are adapted and extended, in an unprecedented way, to handle the specificities of the ultrasound images.