Presentation attack detectionPAD

Detecting when someone holds up a mask, printout, or screen to the camera.

For example, a fraudster might show a photo of someone else's face on a phone screen, and presentation attack detection recognizes that the face on camera is not a live person.

Analyzing a captured biometric sample for the texture, depth, reflectance, or motion cues a live face has that a photo, mask, or screen replay doesn't. The ISO/IEC 30107-3 standard, from the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC), defines how PAD systems get tested. It reports two error rates: the Attack Presentation Classification Error Rate (APCER), how often an attack fools the system, and the Bona Fide Presentation Classification Error Rate (BPCER), how often it wrongly rejects a real, live person.

PAD defends against attacks in which a fake is presented to the capture device, including:

• Printed photos held up to the camera.
• Screen replays, in which a photo or video plays on another device.
• Masks, from paper cutouts to molded silicone or 3D-printed faces.

Attacks that skip the camera and feed fake data straight into the app are injection attacks, and PAD standards do not cover them. Gartner has noted that current PAD testing standards do not address digital injection attacks, which is why services combine PAD with injection attack detection.

Presentation attack detection sits next to terms that are easy to mix up:

• Liveness detection is the name most products use for PAD.
• Spoofing is the attack itself, meaning any attempt to fool a biometric check with a fake.
• Injection attack bypasses the camera, so a presentation attack detector does not see it.

For more, see Incode among top performers in DHS S&T RIVR Track 3 presentation attack detection.