Presentation attack detection (PAD)

A presentation attack is an attempt to deceive a biometric sensor by showing it something other than the genuine trait. The standard's vocabulary calls the fake object a presentation attack instrument, and it covers anything raised in front of the lens as a substitute for a living face.

The cost range is the interesting part of the taxonomy. At one end sits a public photograph and an office printer. At the other sits a mold, a laboratory and a subject who is either cooperating or unaware. A system that reliably defeats the first class and not the second is still worth deploying, provided the organization knows which class it bought.

• A printed photograph of the target's face, flat or curved.
• A photograph or video of the face replayed on a screen.
• A paper cutout mask or a fitted three-dimensional mask.
• A synthetic or deepfake face displayed to the camera.

ISO/IEC 30107-3:2023 governs the testing and reporting of presentation attack detection. It prescribes no detector design and it issues no certificate. What it fixes is the experiment: the artifact species in scope, the number of attempts, the conduct of the operator and the subject, and the form the result has to take.

That is precisely why a result from an accredited laboratory carries weight while a self-assessment does not. Two products can both claim conformance to the same standard and have been tested against very different artifact sets, so the scope of the test is the number worth reading.

Note the phrasing when a vendor cites this. A laboratory tests a specific product version against a specific artifact set at a specific time. Anything stated more broadly than that has left the evidence behind.

Level 1 and Level 2 belong to the iBeta testing program. The standard itself defines no such tiers, which catches out readers who meet the standard number and the level quoted together in a single sentence.

What a tier fixes is the budget and sophistication ceiling for artifacts the laboratory may construct. A higher tier records a larger permitted spend per attack. It does not record a change in the specification, or a product measured against a different one.

Product teams say liveness; the standards say presentation attack detection. A datasheet that chooses the formal term is usually signalling a laboratory result rather than describing a technique, which makes it a useful tell when reading vendor material.

The liveness detection entry approaches the same subject from the operational side: how passive and active checks differ in daily use, and where deepfakes sit in relation to the sensor.

With the face acting as the credential, presentation attack detection is on the critical path of every authentication rather than sitting at its edge. In the two mobile-app methods the enrolled phone performs the capture, the liveness check and the comparison, and a capture that fails never becomes a face token.

The SenseCrypt liveness detection is tested by iBeta to ISO/IEC 30107-3 Levels 1 and 2. Face recognition accuracy is a separate question with separate evidence: the algorithm is evaluated in the NIST Face Recognition Technology Evaluation under Seventh Sense's own name, participation since 2021. The program was known as FRVT when that participation began and now runs as FRTE and FATE. We describe this as NIST-evaluated rather than NIST certified, because NIST evaluates algorithms and certifies nothing. The report card is public: https://pages.nist.gov/frvt/reportcards/11/seventhsense_000.html