Wishart Derived Distance Based Clustering of Polarimetric SAR Images Using Support Vector Machines

Abstract

In this paper, a graph based partitioning procedure for the clustering of fully polarimetric synthetic aperture radar (PolSAR) images is proposed, where support vector machines (SVM) are applied to the Wishart distance that directly is computed from covariance\coherence matrix. The method is a graph based SVM classifier for unsupervised framework. In this schema, the centers of clusters are automatically computed from the average covariance matrices of the final clusters in the refined Freeman decomposition model; and finally graph based SVM is considered for image partitioning. The proposed classifier seizes the advantages of SVM and graph partitioning without suffering from the high computational process that is recognized as a major problem in the graph based classifiers. The performance of the proposed framework using Radarsat-2 and AIRSAR fully polarimetric data is presented and analyzed; and the experimental results show that the framework provides a promising solution for clustering of PolSAR images.