Discriminant Graph Regularized Non-negative Matrix Factorization (DGNMF) for Face Rrecognition

Cai et al develops a graph based approach called Graph Regularized Non-negative Matrix Factorization (GNMF) for parts-based data representation. It should be considered as an unsupervised method since class information is not used in the training set. To take advantage of more information in the training images and improve the performance for classification problem, in this paper, we propose a Discriminant Graph Regularized Non-negative Matrix Factorization (DGNMF) method to discover the manifold structure embedded in high-dimensional face space that is applied for face representation and recognition. In DGNMF, firstly, we encode the geometrical class information by constructing an affinity graph. Secondly, we seek a matrix factorization to respect the graph structure. At last, we do experimental results showing that DGNMF provides a better representation and achieves higher recognition rates in face recognition.