A Multichannel Variational Model for Robust Image Segmentation under Noise

Abstract

This paper presents a novel model for image segmentation under noise: EVRIST (Efficient Variational Representation for Image Segmentation Technique). In order to segment general images containing both quite smooth regions and textures, EVRIST, based on a hierarchical representation obtained by the classical weak membrane, utilizes two channels: one relative to the mean and another for the edges density. The results achieved on both manual compositions and on real images, show the high robustness to the noise along with a low computational effort.