Low-Resolution Face Recognition System Using Siamese Network

Practically face recognition is applied to a not really high-resolution face image due to hardware limitation such as surveillance camera or low-quality webcam. Low resolution face recognition system is a system consists of several key components which are face detection, super resolution, and face recognition. Siamese network is useful to be implemented as a part of face recognition component for recognizing unbounded face identity and doing one shot learning to record newly recognized identity. In this paper, there are three main goals to be achieved in the end of the research which are: 1) developed a complete low resolution face recognition system, 2) compared quantitatively between every techniques for each component, 3) using the best technique to be implemented for each component in the system. We have successfully achieved 93% accuracy for 36×36 face image, 84% for 24×24, and 56,2% for 12×12 with 0.086s execution time.