Abstract
As a sub-task of face recognition in the field of computer vision, near-infrared (NIR) and visible light (VIS) face recognition has been widely assessed for the capacity to match face images between NIR modality and VIS modality. NIR-VIS face recognition techniques hold great potential to enable facial identification in unconstrained environments across a diverse suite of applications, such as law enforcement, surveillance and military. Nonetheless, the modality variation between NIR and VIS gives rise to different facial appearances, imposing a major challenge in face recognition to be addressed. This paper reviews datasets and state-of-the-art works in the field of NIR-VIS face recognition, which can be divided into three categories: image synthesis-based methods, subspace learning-based methods and invariant feature-based methods. In addition, this paper provides an analysis of the contributions made by, and potential improvements to, each of these three approaches.