Interactive gestures reveal real-time face swaps better than expressions
Not All Challenges Are Equal: Evaluating Interactive Liveness Detection Against Real-Time Face Swaps
Computer Vision and Pattern RecognitionCryptography and Security
Summary
It’s important to know if the person in front of a camera is real, but new tools can swap faces in real time and trick systems. The authors tested different types of interactive challenges, like head movements and facial expressions, to see which best catch these face swaps. They found that asking for head poses or combined gestures works better than just facial expressions. Also, how long a person takes to respond can help spot fake video too.
What this means in practice
- •For identity verification services: Improve detection of real-time face swaps during online ID checks using randomized head pose and gesture challenges.$Commercial implications: Enables more reliable fraud prevention products for companies verifying identities remotely against real-time deepfake attacks.
- •For security system developers: Design liveness tests leveraging response timing and specific gestures to better distinguish live users from manipulated camera feeds.
Authors
Anton Firc, David Drtil, Martin Perešíni, Milan Šalko, Vojtěch Staněk, Kamil Malinka
Abstract
Remote identity verification assumes that a camera stream shows a live person, yet public real-time face-swap tools can replace that person before the stream reaches the verifier. Passive deepfake detectors generalize poorly to new generators, while interactive challenges are often selected without evidence of which gestures expose current attacks. We evaluate randomized gesture challenges in 108 live sessions with six subjects, 24 impersonated identities, and two face-swap tools, DeepFaceLive and FaceFusion, injected through a virtual camera. The static calibration segment reaches a session-level AUC of 0.878 with a frozen UCF spatial detector, while the mean across the active-gesture windows reaches 0.988. Pose and combined gestures have the highest point estimates, whereas expressions show no clear improvement. Hand occlusion against FaceFusion, configured with an occlusion mask, scores below the static baseline. The end-to-end time required to satisfy the validator also distinguishes attacks from genuine sessions, achieving a session-level AUC of 0.732, although this signal includes synthesis and transport delays. These results show that challenge sets should be tested against current attack configurations and that response time can provide an additional, deployment-specific signal.