Clustering high-dimensional data via spectral clustering using collaborative representation coefficients

Abstract

Clustering high-dimensional data is challenging for traditional clustering methods. Spectral clustering is one of the most popular methods to cluster high-dimensional data, in which the similarity matrix plays an important role. Recently, sparse representation coefficients have been proposed to construct the similarity matrix via the cosine similarity between each pair of coefficient vectors for spectral clustering and showed promising results. However, the sparse representation emphasizes too much on the role of ℓ1-norm sparsity and ignores the role of collaborative representation, which makes its computational cost very high. In this paper, we propose a spectral clustering method based on the similarity matrix which is constructed based on the collaborative representation coefficient vectors. Extensive experiments show that the proposed method has a strong competitiveness both in terms of computational cost and clustering performance.