Single Sample Face Recognition: A Comparative Study between Texture and Deep-Learning-based Features

Single sample face recognition (SSFR) is an active and challenging research subject in the biometric and artificial intelligence communities with the ultimate objective of recognizing individuals using only one facial image per class in the training set. Arguably, feature extraction is the crucial phase in constructing any pattern recognition system, which aims to extract the relevant information that characterizes each pattern. Texture and deep-learning-based approaches are the most prominent strategies for extracting relevant information. This paper compares several SSFR models constructed with texture and deep-learning-based features. We tested four common texture descriptors and seven pre-trained deep learning-based convolutional neural networks (CNNs) models to extract the discriminating face features. Finally, based on the K-Nearest Neighbor (K-NN) classifier, we performed the classification process. The results of the experiments on the unconstrained Alex & Robert (AR) dataset showed superior performance using texture-based approaches compared to the deep-learning-based strategies.