Sampled FLDA for face recognition with single training image per person

The Fisherface is one of the most successful face recognition methods, which however, cannot be directly applied to face recognition where only one sample image per person is available for training. In this paper, a method is proposed to obtain multiple training samples from a single face image by sampling, and then Fisher linear discriminant analysis (FLDA) is applied to the set of newly produced samples. Experimental results on the ORL face database show that the proposed method is feasible and has higher recognition performance than E(PC)²A and SVD perturbation algorithms.