What Is a Presentation Attack?
A presentation attack is an attempt to spoof a face recognition system by presenting an artificial representation of a face rather than a live face. Common attacks include: printed photo attacks (holding a photo of the target in front of the camera), screen replay attacks (showing the target's face on a phone or tablet), 3D mask attacks (wearing a 3D-printed or sculpted mask), and deepfake video attacks (using AI-generated video).
Face recognition systems without anti-spoofing are vulnerable to these attacks. A system that only checks whether the face matches a database entry, without checking whether a live face is present, can be defeated by presenting the correct photo.
How Liveness Detection Works
Multiple technical approaches to liveness detection exist. Texture analysis detects the characteristic flat texture and rectilinear artefacts of printed or displayed images. 3D depth sensing (used in iPhone Face ID) confirms the face is a three-dimensional object at the expected depth. Motion-based approaches require the user to perform specific movements (blink, turn head) that are difficult to replicate in a replay attack. Passive video analysis detects rPPG (remote photoplethysmography), the subtle colour changes in skin caused by blood pulsation, which cannot be replicated by a static photo.
For Ollie's celebrity matching application, presentation attack vulnerability is less concerning than for identity verification applications, the attack incentive is lower. But as AI face generation improves, detecting AI-generated input will become relevant for maintaining database integrity.