A Hybrid Active Contour Model based on New Edge-Stop Functions for Image Segmentation

Abstract

Edge-based active contour methods are popular algorithms for image segmentation, with the purpose to extract the area of interest. However, they may face to boundary leakage and improper segmentation when handle images under weak edges or complex shapes. The extensive edge-stop functions adopt edge information, which cannot apply to guide the evolving curve approaching to target boundaries. To resolve this issue, a novel level set algorithm based on non-local means (NLM) filtering is constructed in this study. Firstly, the images are subjected to non-local means filtering to generate edge map. Secondly, a new edge-stop function constructed from this edge map as well as the fuzzy k-NN classification algorithm is incorporated into the variational model. Our experiments demonstrate that non-local means filtering is able to sharp edges both on medical and natural images. Thus, this analysis seems to be useful for clinical medical diagnosis.