Cooperative Synthetic Aperture Radar Image Segmentation Using Learning Sparse Representation Based Clustering Scheme

Based on a recent proposed and popular sparse representation based classifier (SRC), in this paper we presented a novel Learning Sparse Representation based Clustering (LSRC) scheme for Synthetic Aperture Radar (SAR) segmentation. LSRC introduces the examples-based dictionary learning technology in SRC to find a dictionary that is adaptable to sparsely representing samples, which is liable to provide more accurate approximation of samples and subsequently achieve higher classification accuracy rate. Moreover, for the intrinsic supervised nature of LSRC, we adopt an unsupervised-clustering cooperative approach to provide training samples for LSRC, in which some "good" samples with higher membership degrees are selected from the clustering result of K-means algorithm. Some experiments are taken on segmentation of both the texture images and SAR images to investigate the performance of our proposed method, and the results prove its superiority to its counterparts.