A fast segmentation algorithm with curvature-independent direction based on the Chan-Vese model

To improve image segmentation accuracy with better edge details, a new fast method is proposed based on the Chan-Vese(C-V) model. It combines an edge function and a signed distance function. The edge function is directionally curvature-independent, and the energy function evolves without re-initializing the signed distance function. The improved method extends the C-V model, so as to properly extract contours from given images in homogeneous areas. It does not use the local gradient information of level sets while evolving contours, instead it adds a curvature-independent directional edge function and uses mean curvature motion to minimize length energy. The internal energy function term of the energy function is increased to simplify and speed up the model when it needs to re-initialize the signed distance function. Experiments show that the new algorithm nicely evolves wanted target edge contours for accurate image segmentation, and also reduces time significantly, approximately 1.2 times faster than the geometric active contour C-V model. © 2015, Editorial Board of Journal of Harbin Engineering. All right reserved.