Low-Resolution Face Recognition Using Multi-Stream CNN in Siamese Framework

Face recognition systems have been significantly advanced in recent times by the emergence and advancement of deep learning approaches. However, low-resolution face recognition is still a challenging task, especially when the images are snapped in an unconstrained condition. Such conditions are common in surveillance applications, where the snapped facial images often contain blur, non-uniform illumination, non-frontal pose, and occlusions. Further, the resolution of the images is often restricted due to different constraints. In this paper, we propose a network architecture consisting of multi-stream CNN based on the Siamese framework for low-resolution face recognition. Multiple streams of the CNN extracts complementary discriminative information from face images suitable for recognition purpose. The proposed architecture is evaluated on SCface, LFW, and YouTube face datasets.