Face Recognition Based on Non-Negative Factorization and FLDA for Single Training Image per Person

Abstract: Dimensionality reduction is performed by both Principal Component Analysis (PCA) and Fisher Linear Discriminant Analysis (FLDA). Covariance matrix and Eigen vector approach is followed in PCA. FLDA finds within class and between class scatter matrices. In some situations, within class scatter matrix may become singular. Normally singular matrix does not have inverse. Two or more virtual samples are generated from the training set to avoid this problem. A new method called non-negative matrix factorization is used in single image per person problems. The proposed method performs better than SVD, QRCP and SDD method in terms of recognition rate. But training time is slightly more than QRCP and better than SVD and SDD approach.