Facial marks : detection and integration into a facial recognition system

Facial marks can be used as a complement to facial recognition systems, and have been used for this purpose before. New advances in convolutional neural network (CNN) architectures have enabled more accurate detection of facial marks. In this paper state-of-the-art CNN models are trained on the FRGCv2 dataset for the recognition and detection of facial marks. The resulting systems are combined with a FaceNet facial recognition system in order to improve facial recognition performance. In particular, the improved ability to differentiate between twins will be studied. Due to their similarities, twins are exceptionally difficult for facial recognition systems to distinguish. The Twins Days dataset is used in order to investigate the ability of the combined system to differentiate between twins. This paper demonstrates significant improvements in facial recognition and twin differentiation performance when facial mark detection is used.