Face Anti-Spoofing by Learning Polarization Cues in a Real-World Scenario

Face anti-spoofing is the key to preventing security breaches in biometric recognition applications. Existing face anti-spoofing approaches usually well recognize the spoofing attacks when testing in particular datasets. But the performance drops drastically when it comes to the actual scene. In this paper, we try to boost the generalizability capability by learning the polarization features of human faces in real-time. A human face anti-spoofing method suitable for real scenario has been proposed, which resists spoofing attacks by automatically learning the physical characteristics of polarized biometric images. A computational framework is developed to extract and classify the unique face polarized features using convolutional neural networks and SVM together. Extensive experiments demonstrate the adv antages of our real-time polarized face anti-spoofing (PAAS) technique to counter diverse face spoofing attacks (print, replay, mask) in uncontrolled indoor and outdoor conditions after learning the polarized face information of 108 people. A four-directional polarized face image dataset (CASIA-DOLP) is released to inspire future applications within biometric anti-spoofing field.