Diffusion-Based Hybrid Level Set Method for Complex Image Segmentation

Abstract

To capture the weak boundary in complex image, we proposed a new level set method. Different from the exiting methods, our method is performed on diffusion space rather than intensity space. The total energy functional is a linear combination of local part, global part and regularization part. Firstly, the nonlinear diffusion is performed on intensity image to acquire diffused image. Then, the local energy term is formed by implementing a local piecewise constant search on diffused image. To further avoid local minimum, the global energy term is constructed by approximating diffused image in a global piecewise constant way. Besides, the regularization energy term is included to naturally force level set function to be signed distance function. Finally, image segmentation can be performed by minimizing the overall energy functional. The experiments on several complex images with distinct characteristics have shown the powerful boundary approaching ability of our method.