Face Identification System using Deep Learning Method at Low Lighting Resources

The topic of identifying people in low light using near infrared cameras has recently attracted increased attention in biometric systems due to its benefits of the less sensitive to illumination levels. Currently, deep learning facial recognition models provide the best accuracy results for various databases of visible faces. Unfortunately, the desired performance cannot be achieved using only pre-trained deep learning facial recognition models for near infrared image databases. In this paper, we present a NIR-VIS image translation model which is available to translate from near infrared facial images into their corresponding visible images. The evaluation performances are presented by applying the pre-trained face embedding model on two different NIR databases of facial images.