Face Recognition From Single Sample Per Person by Learning of Generic Discriminant Vectors

The conventional ways of recognizing faces always assume the possession and heavily relies on extensive and representative datasets, but that is not the case in most real-world situations where more often than not, a very limited or even only single sample per person (SSPP) is available which ultimately rendering most face recognition systems to fail severely. This paper proposes a development of face recognition based on a combination of traditional eigenface with artificial neural network (ANN), having the face recognition performance boosted by the classification of discriminant vectors learned from a set of generic samples. The discriminant vectors representing intra-subject and inter-subject variations are learned based on similarities of pairs of generic samples which then used to classify novel intra-subject pairs and inter-subject pairs from probe set and corresponding gallery set. After that, the resulting classification is used to recognize faces by combining it with the expressive ability of eigenface via a voting procedure. The proposed method when tested with FERET and YALE datasets suggests that in face recognition within the SSPP constraints, the performance of the proposed method is better than some state-of-the-art methods.