Detection of Presentation Attacks in Face Recognition Systems with Non-contact Photoplethysmography

Abstract

Presentation attack detection (PAD) is a challenging task for biometric identity verification systems based on face recognition. A presentation attack could be done using a 3D mask, a printed static image obtained from a laser or inkjet printer, or a video replay to exploit the vulnerabilities of the target face recognition system. In order to detect presentation attacks, features, which are useful for liveness detection could be used as clues. Non-contact (remote) photoplethysmography (rPPG) methods could be used to estimate physiological signals such as the heart rate, respiratory rate and heart rate variability that indicate liveness features using only a camera recording without any additional hardware. These physiological signals could be used as features to train a machine learning classifier to differentiate fake faces from genuine faces. In this study, we first provide a literature survey of presentation attack detection methods using non-contact photoplethysmography. We also provide an experimental comparison of PAD using different PPG methods on the 3DMAD dataset.