Virtual Label Constraint Nonnegative Matrix Factorization

Abstract

This paper proposes a novel Semi-supervised Nonnegative Matrix Factorization (NMF), called Virtual Label Constraint Nonnegative Matrix Factorization (VLCNMF). The idea of the VLCNMF is to extend the NMF by incorporating a virtual label constraint into the NMF decomposition. Different from previous works, our approach covers two main steps: the first step is to obtain virtual labels by label propagation algorithms and the second step is to add these virtual labels information as additional constraints into original NMF. The proposed VLCNMF approach is applied to the problem of semi-supervised image representation using the well-known ORL, Yale datasets.