Adaptive Distance Regularized Level Set Method and Its Application to Image Segmentation

The distance regularized level set method has the advantage of maintaining the regularity of the level set function without re-initialization. However, it has the disadvantage of requiring the initial curve around or inside the detected objects. In this paper, an adaptive distance regularized level set method is designed. Firstly, the local energy term is introduced to make the method robust to how and where the initial curve is selected and can successfully segment the images with intensity heterogeneity. Secondly, a Gaussian filter is utilized to ensure the smoothness and regularity of the level set function and eliminate re-initialization. Furthermore, the narrow band method is applied in the adaptive distance regularized level set method to reduce the computation and the convergence time. The results of the comparative experiments show the advantages of the designed method.