Improved Chan-Vese image segmentation model using Delta-Bar-Delta algorithm

Abstract

The level set based Chan-Vese algorithm primarily uses region information for successive evolutions of active contours of concern towards the object of interest and, in the process, aims to minimize the fitness energy functional associated with. Orthodox gradient descent methods have been popular in solving such optimization problems but they suffer from the lacuna of getting stuck in local minima and often demand a prohibited time to converge. This work presents a Chan-Vese model with a modified gradient descent search procedure, called the Delta-Bar-Delta learning algorithm, which helps to achieve reduced sensitivity for local minima and can achieve increased convergence rate. Simulation results show that the proposed search algorithm in conjunction with the Chan-Vese model outperforms traditional gradient descent and recently proposed other adaptation algorithms in this context. © Springer International Publishing Switzerland 2014.