A Practical Design for Face Recognition with Anti-Spoofing Based on Non-Visible Light Cameras

Abstract

As one of the most promising artificial intelligence technologies, face recognition technology has been traditionally associated with security and has recently expanded into other industries such as retail, marketing, health, etc. A challenging issue for face recognition is face spoofing where imposters use a variety of fake faces in an attempt to deceive the face recognition systems. Extensive research has been conducted on face anti-spoofing algorithms and systems. Some researchers tend to design highly advanced and thus complicated algorithms assuming that regular visible light cameras are used so as to reduce peripheral hardware cost. However, implementation of such complicated algorithms may need support of quite advanced processing hardware which costs even much more than what has been saved by the cameras. Furthermore, such designs are not suitable for some practical application scenarios where the space is compact and small distributive devices without super computation strength and with low power consumption have to be used. Complicated algorithms also come with decrease in stability and generalization. An effective solution to reduce algorithms complexity is to capture images using thermal cameras which are inherently convenient for detecting various presentation attacks (PA). Sole rely on thermal cameras would need very high resolution thermal lens and thus bring the problem of much increased hardware cost. Based on extensive study on existing related work and electromagnetic theories and our experience in related practical projects and products, we proposed a practical design for face recognition with anti-spoofing. In our design, a low resolution thermal camera works together with a near infrared camera for face anti-spoofing and recognition. This design is believed to fight PA effectively and recognize faces accurately with relatively low cost and compact devices. These features add to our design's competitive strength in practical applications.