GENERALIZED CHAN-VESE MODEL FOR IMAGE SEGMENTATION WITH MULTIPLE REGIONS

摘要

In this paper we propose a modified region-based active contour model by integrating the local information of foreground region into Chan-Vese model. By adding a local fitting term into the conventional energy function, our model is able to overcome two limitations of the previous region-based level set methods: high sensitivity to the location of initial contours and inability to segment multiple objects. We will show the advantages of our method compared with other major level set-based techniques in terms of both efficiency and accuracy by segmentation experiments.