An active contour for segmentation of images of low contrast and blurred boundaries

Abstract

A novel level set-based active contour model (LSAC) composed by region and boundary terms is proposed to segment the images featured by low contrast and blurred boundaries. The region terms derived from weighted cross-entropy play major role to locate object boundary and the boundary term derived from direct detection of image gradient plays supplementary role for promotion of segmentation accuracy. Moreover, the numeric method used provides good numeric accuracy. The Experimental results show the model exactly locates blurred boundaries between adjacent image regions that have highly similar intensities.