Unsupervised face anti-spoofing using dual cameras based feature matching

Face anti-spoofing is a crucial part of face recognition system to protect subject's privacy and life safety. Most current face anti-spoofing algorithms are based on feature extraction and machine learning. The performance of machine learning based approaches depends on the quantity and quality of the training data. In this paper, we propose an unsupervised face anti-spoofing method based on feature extraction and matching of a dual camera setup, which does not require offline training. The principle of our method is simple, intuitive, and generally applicable. The core idea of our method is exploiting the fact that a 3D face has different feature representations in images from two cameras with different view angles, as compared to that of a 2D spoofing face (either printed in a paper or showing on a screen). The proposed method has been benchmarked on a dataset created by our dual camera setup and shows an accuracy of 94.2%.