Active Contours Driven by Kernel-Based Fitting Energy

In this paper, a new region-based active contour model using kernel-based fitting energy is proposed to improve the accuracy and efficiency of segmentation. The proposed kernel-based fitting energy is defined as a kernel function inducing a robust non-Euclidean distance measurement to segment images more effectively. In addition, an exponential-type kernel-based function in our model is used, which leads to faster converge. At last, to avoid costly computation of re-initialization widely adopted in traditional level set methods, we introduce a new penalty energy as a regularization term. Experimental results demonstrate that our model can segment images more precisely and much faster than the well-known Chan-Vese model.