A new constraint spectral clustering algorithm

Abstract

Subspace clustering methods based on spectral clustering have been very popular due to their theoretical guarantees and empirical success. However, considering the constraint information of data, these subspace-clustering-based constraint clustering algorithms are difficult for the high-dimensional data with data nuisances to achieve better clustering results. This paper proposes a novel constraint spectral clustering algorithm based on the program of sparse subspace. Firstly, constraint term which are suitable for sparse subspace model are established according to the different statuses of representation matrix. Then a novel semi-supervised sparse subspace model is presented with the constraint terms mentioned above. Finally, the final clustering results could be acquired by spectral clustering under the guidance of constraint information. Experiments on two real-world dataset verify the property of the algorithm and show that this approach could achieve better clustering accuracy than others.