PulseEdit: Editing Physiological Signals in Facial Videos for Privacy Protection

Recent studies have shown that physiological signals such as heart beat and breathing can be remotely captured from human faces using a regular color camera under ambient light. This technology, referred to remote photoplethysmography (rPPG), used collect the status of users who are in front camera, which may raise privacy concerns. To avoid abuse rPPG this paper develops PulseEdit, novel efficient algorithm edit facial videos without affecting visual appearance thus protect user’s signal disclosure. PulseEdit either remove trace video or transform contain target chosen by user. Experimental results show effectively prevent rate measurement based on rPPG. It is possible utilize adversarial scenarios against rPPG-based security algorithms. We present analyses performance liveness detection deepfake detection, demonstrate its ability circumvent these algorithms important role supporting design attack-resilient systems.