An Improved GAN-Based Method for Low Resolution Face Recognition

Abstract

The Low Resolution Face Recognition (LR FR) issue focuses on the resolution variation problem mainly displayed in surveillance systems where faces are captured in an unconstrained environment. To address this challenging problem, we introduced a new method to recognize faces under Low and Very Low Resolution conditions. The proposed method consists of two phases : an off-line phase and an inference phase. In the off-line phase, we generate a model to transform Low-Resolution (LR) face images into High-Resolution (HR) face images via a Generative Adversarial Network (GAN). As for the inference phase, we make use of the already generated model to improve the face resolution then we extract deep features to identify the face. An experimental study was carried out on the famous ORL and FERET databases and the obtained results proved the efficiency of the proposed method to deal with Low Resolution (LR) and Very Low Resolution (VLR) Face Recognition problem.