Constrained Graph Concept Factorization for image clustering

Matrix factorization techniques have been frequently applied in data representation and pattern recognition. One of them is Concept Factorization (CF), which is a new matrix decomposition technique for data representation. In this paper, we propose a novel semi-supervised matrix factorization algorithm, called Constrained Graph Concept Factorization (CGCF), which incorporates the label information as additional constraints. Specifically, CGCF preserves the intrinsic geometry of data as regularized term and use the label information as semi-supervised learning, it makes nearby samples with the same class-label are more compact, and nearby classes are separated. An efficient multiplicative updating procedure was produced along with its theoretic justification of the algorithmic convergence. Compared with NMF, GNMF, CF, LCCF and Kmeans, experiment results on ORL and YALE face databases have shown that the proposed method achieves better clustering results.