Multiphase segmentation of SAR images with level set evolution

Segmentation is a fundamental problem for the automatic interpretation of synthetic aperture radar (SAR) images. We propose an improved multiphase segmentation method for SAR images based on Chan-Vese level set evolution model. The costly re-initialization procedure of signed distance function in Chan-Vese model is eliminated through introducing a penalty term. The proposed method has two advantages over traditional multiphase segmentation scheme with level set evolution. First, regions with close gray scale could be segmented into different class correctly by utilizing several binary segmentations of irregular regions. Second, falsely segmented fractions with one pixel width on the edges of uniform regions are eliminated with morphological open operation. The numerical algorithm using finite differences is also presented, which has been applied to simulated images and real SAR images with more promising results.