Breaking Infrared Recapture Detection: Optical-Synthesis Attacks and Depth-Aware...
summary
An attack called SynthIR uses an inexpensive hot mirror to split visible and IR paths so the RGB camera sees a monitor image while the IR depth sensor sees a physical 3D proxy; attackers build low‑cost cardboard objects or generate images aligned to an IR depth map with ControlNet (and animate with SadTalker) and achieve 100% attack success in experiments including video and printed images. A proposed dual-pixel detection pipeline estimates stereo depth from the same optical spectrum and flags all recaptures with 100% accuracy across Pixel/Canon/iPhone datasets.
AI-written summary · about
chapters (14)
- 0:00C2PA and re-capture attacks
- 1:38Two-sensor depth verification flaw
- 2:22Bypassing two-sensor defense
- 2:44SynthIR and hot mirror technique
- 3:49Threat model and attacker abilities
- 4:33Scene Splicer with hot mirror
- 5:16Image-first attack success
- 6:23iPhone depth sensor fusion flaw
- 7:29Depth-first approach with ControlNet
- 8:51Video attack with SadTalker
- 9:14Defense using dual pixel sensors
- 10:16Recapture detection with DP sensors
- 11:19Evaluation results 100% accuracy
- 12:27SynIR attack and defense summary
related talks
14mRed Bleed: A Pragmatic Near-Infrared Presentation Attack on Facial Biometric...
30mInfrared: Old Threat Meets New Devices
16mPerspective-Shift Attacks Against Optical Perception Sensors: A Novel Attack...
01h 06mSelf-Driving and Connected Cars: Fooling Sensors and Tracking Drivers
17mScoop: Mitigation of Recapture Attacks on Provenance-Based...
28mFace spoofing detection: theory and practice
10mInvestigating Stealthy Latency Attacks Against Camera-Based Object...
26mVirtual U: Defeating Face Liveness Detection by Building Virtual Models...