SAR image superpixels by minimizing a statistical model and ratio of mean intensity based energy

Abstract

Superpixel based SAR image classification methods can take advantage of the contextual information in SAR images effectively, leading to robust classification results. The accuracy of superpixel generation has great impact on the performance of the following classification stage. In this paper, based on the property of SAR images, an energy minimizing based superpixel generation approach is proposed for SAR images. The energy function is composed of two parts. The data term is defined according to the statistical characteristic of SAR images, and the regularization term is defined by using the ratio of mean intensity. Then the superpixel generation is performed by energy minimizing with graph cut based energy minimization method. Experimental results on both synthetic and real SAR image data verify the good performance of the proposed approach. Compared with several superpixel approaches, the proposed approach can deal with speckle noise more effectively, resulting in better applicability for SAR images.