Hybrid diffusion-steered model for suppressing multiplicative noise in ultrasonograms

Abstract

Ultrasonograms refer to images generated through ultrasonography, a technique that applies ultrasound pulses to delineate internal structures of the body. Despite being useful in medicine, ultrasonograms usually suffer from multiplicative noises that may limit doctors to analyse and interpret them. Attempts to address the challenge have been made from previous works, but denoising ultrasonograms while preserving semantic features remains an open-ended problem. In this work, we have proposed a diffusion-steered model that gives an effective interplay between total variation and Perona-Malik models. Two parameters have been introduced into the framework to convexify our energy functional. Also, to deal with multiplicative noise, we have incorporated a log-based prior into the framework. Empirical results show that the proposed method generates sharper and detailed images. Even more importantly, our framework can be evolved over a longer time without smudging critical image features.