Mumford-Shah level set method for multi-objective contour extraction
Kun Liu · Caai Transactions on Intelligent Systems · 2011
Fast detection of objective contours and extraction of its geometric shape have important roles in graphics and image processing.Based on the Mumford-Shah model,a novel level set method for multi-objective contour extraction was presented.First,the gradient vector field was combined with normal direction of the curves as boundary abstracted fields,so as to generate a bi-directional geometric deformable flow field which can drive active contours evolving towards the boundary from inside or outside edges.Furthermore,the distributed information of the image would be left as area evolution energy.This method can solve problems that arise when area energy information is lost because local geometric information isn't considered,or when topological structure should not be changed because the gradient vector field is orthogonal with normal direction.Then the level set function was modified so that it could change adaptively in curve convergence.Other reasons for this modification were to make sure that the level set changes could maintain signal distance function,the search area could be covered sufficiently before re-initialization,and the iterative number could be decreased.The convergence efficiency was also raised.Finally,a numerical solving scheme was given.Experimental results illustrate that the method proposed in this paper is feasible and robust.