SAR image segmentation with combined Bayes’ rule, GMTRJ, and EM algorithms

Abstract

ABSTRACT This study analyses the roles of all the classes in the segmentation procedure, and aims to improve the segmentation accuracy of synthetic aperture radar (SAR) images. Accordingly, weight variables of all classes for SAR images are defined and incorporated into the segmentation model by Bayes’ rule and geometric partition tessellation. In addition, the Generalized Multiple-Try Reversible Jump (GMTRJ) and expectation maximization (EM) algorithms are designed to obtain the values of weight variables and the optimal segmentation by simulating the segmentation model. The greater the value of the weight variable, the more important the role of its class. The proposed approach can automatically determine the classes’ roles in the segmentation procedure and segment SAR images accurately and quickly. Finally, the proposed and comparison approaches are tested on SAR images, and the results fully demonstrate the effectiveness of the proposed approach.