Global-to-local region-based indicator embedded in edge-based level set model for segmentation

Abstract

Image segmentation is an essential analysis tool in the field of computer vision, and the level set method has been widely used in image segmentation. Specifically, the edge-based level set models can reduce many undesired regions because they mainly rely on the edge information. However, the edge-based level set models are usually sensitive to the initial condition, which limits their application. To overcome this shortcoming, a global-to-local region-based indicator is designed in this paper, which is utilized to embed the region information into the edge-based models. Unlike the edge-based indicator frequently used in the edge-based models, the proposed region-based indicator can allow bidirectional motion of the active contour curve according to the region information. In general, the proposed region-based indicator can intrinsically incorporate the edge information and region information into one single energy function. Experimental results on synthetic images, natural images and medical images validate the effectiveness of the proposed method. Compared with some other level set models, the proposed method generally achieves better performance.