An improved distance regularized level set evolution without re-initialization

Abstract

Level set methods have been widely used in image processing and computer vision. The re-initialization problem of level set limits its application. Recently proposed distance regularized level set evolution (DRLSE) can avoid level set re-initializations, the DRLSE formulation allows the use of more general and efficient initialization of the level set function and provides a simple narrowband implementation to greatly reduce computational cost. However the diffusion rate may incur undesirable side effect in some circumstances, and thus influence the distance regularization. An improved diffusion rate model is proposed in this paper, and experiment results show that our model performs better in distance regularization, and moreover the example of applying our model in image segmentation task indicates it has more widely applications in other image processing tasks.