Gradient Descent Fisher Non-negative Matrix Factorization for Face Recognition
Yuhao Liu · Journal of Information and Computational Science · 2013
In this paper, we propose a novel subspace method called Gradient descent Fisher Non-Negative Matrix Factorization (GdFNMF) for face recognition. By imposing fisher constraints and taking use of Euclidean distance as the measure of the cost function into gradient descent method, our GdFNMF can encode discrimination information for the classification problem. Experiments show that our GdFNMF achieves better performance than FNMF and less sensitive to the value of parameter.