Graph-regularized CF with local coordinate for image representation

Abstract

Concept factorization (CF) has been a powerful data representation method, which has been widely applied in image processing and document clustering. However, traditional CF cannot guarantee the decomposition results of CF to be sparse in theory and do not consider the geometric structure of the databases. In this paper, we propose a graph-regularized CF with local coordinate (LGCF) method, which enforces the learned coefficients to be sparse by using the local coordinate constraint meanwhile preserving the intrinsic geometric structure of the data space by incorporating graph regularization. An iterative optimization method is also proposed to solve the objective function of LGCF. By comparing with the state-of-the-arts algorithms (Kmeans, NMF, CF, LCCF, LCF), experimental results on four popular databases show that the proposed LGCF method has better performance in terms of average accuracy and mutual information.