Segmentation for SAR Image Based on a New Spectral Clustering Algorithm

Abstract

A new spectral clustering (SC) algorithm with Nystrom method is proposed for SAR image segmentation in this paper. The proposed algorithm differs from previous approaches in that not only with Nystrom method are employed for alleviating the computational and storage burdens of the SC algorithm, but also a new similarity function is constructed by combining the pixel value and the spatial location of each pixel to depict the intrinsic structure of the original SAR image better. Our algorithm and the classic spectral clustering algorithm with Nystrom method are evaluated using the real-world SAR images. The results demonstrate the running time and the error rate of the proposed approach and the classic spectral clustering algorithm with Nystrom method.