Similarity matrix construction methods in sparse subspace clustering algorithm for hyperspectral imagery clustering

Abstract

The clustering of hyperspectral images is a challenging task because of the high dimensionality of the data. The sparse subspace clustering (SSC) algorithm is one of the popular used clustering algorithm for high dimensionality data. But, SSC has not considered the spectral and spatial information fully, so it is not satisfied for Hyperspectral Imagrery (HSI) clustering. In this paper, a novel similarity matrix construction methods are proposed which combined the high spectral correlation and rich spatial connection. Firstly, we utilize the cosine similarity of sparse representation vector to construct a novel similarity matrix. Then, the similarity matrix based on Euclidean distance of the sparse representation vector can connect spectral correlation with spatial information. Several experiments on HSIs demonstrated that the proposed algorithms are effective for hyperspectral images (HSIs) clustering.