A Land Cover Mapping Algorithm Based on a Level Set Method

Abstract

A novel supervised classification algorithm is presented for remotely sensed images using the level set method under a statistical framework. The level set method was employed to capture the connectivity properties of land cover classes. This work demonstrated that land cover mapping under the maximum a posteriori criteria can be converted into an energy minimization problem of level set functions. Since the level set functions are real-valued, the optimum solution can be easily obtained from a gradient search technique. The experimental results showed significant improvements in term of the classification performance of the approach on both synthetic and satellite images when compared to the maximum likelihood classifier.