Recognizing faces in low-light and nighttime conditions is a challenging problem due to the noisy and poor quality nature of the images.Thermal imaging is often used to obtain facial biometric in such conditions. Thermal face images, while having a strong signature at nighttime, are not typically maintained in biometric-enabled watch lists and so must be compared with visible-light face images to enable face recognition in low lighting conditions.In this project, wea thorough feasibility study to investigate the limits of current deep learning-based cross-modal matching algorithms for visible-to-thermal matching.In particular, the goals of this project are as follows:-Collect a dataset consisting of face signatures across the spectrum from 100 subjects using the Army Research Laboratory Government Furnished Equipment (GFE).-Integrate, fine-tune and evaluate the best performing face verification and recognition system from the JANUS project on the new heterogeneous face dataset.-Design and evaluate a robust deep learning-based domain adaptive matching system (or systems) for cross-domain face recognition. Address the risks and potential payoffs of this technology
