HDLHC: Hybrid Face Anti-Spoofing Method Concatenating Deep Learning and Hand-Crafted Features

Despite significant attention given to face spoofing, there is still a need for more generalizable face anti-spoofing methods that would perform robustly in practical face recognition systems. Face spoofing attacks can be done by presenting a photo, video or a mask of the target person to the camera. This paper introduces a feature level fusion method, called HDLHC, that concatenates features extracted automatically by deep learning with hand-crafted image quality features derived from original images. Extensive experiments demonstrate that HDLHC outperforms the state-of-the-art methods on the Oulu-NPU and SiW datasets. It demonstrates its generalization ability under different face spoof attack conditions.