The Improved CESSC Algorithm Based on Meanshift Sparse Subspace Clustering for Hyperspectral Images

Abstract

Hyperspectral images are rich in spatial information and spectral information. And hyperspectral image (HSI) clustering is very important for the study of HSI. However, just like all pixel-based clustering methods, traditional sparse subspace clustering(SSC) algorithm also faces these problems: (1) the effect of salt-and-pepper noise on the experimental results is more prominent; (2) it can not make full use of the different neighborhood relationships; (3) it result in high computational complexity. In this paper, we have improved the Cosine-Euclidean similarity matrix(abbreviated as CE) Construction Based on Sparse Subspace Clustering(CESSC) algorithm for Hyperspectral images to solve these three problems. First of all, we use the Meanshift algorithm to segment the original hyperspectral data, so, we get a series of regions and use the mass center of each region to represent all points in this region, then, the algorithm of Cosine-Euclidean Similarity Matrix Construction Based on Sparse Subspace Clustering is conducted on all this center of mass to get clustering labels; finally the else points in each region obtain their final clustering result by giving the same labels with their center of mass. Experimental results show that the proposed algorithm can improve the recognition accuracy to some extent.