A Hierarchical Student's t-Distributions Based Unsupervised SAR Image Segmentation Method

摘要

We introduce a finite mixture mode using hierarchical Student's distributions, called hierarchical Student's t-mixture model (HSMM), for SAR images segmentation. The main advantages of the proposed method are as follows: first, in HSMM, the clustering problem is reformulated as a set of sub-clustering problems each of which can be solved by the traditional SMM algorithm. Second, a novel image content-adaptive mean template is introduced into HSMM to increase its robustness. Third, an expectation maximization algorithm is utilized for HSMM parameters estimation. Finally, experiments show that the HSMM is effective and robust.