Ayonix Face Recognition

In short

Liveness detection assesses whether a presented face comes from a live person rather than a photograph, screen replay or mask. A liveness claim means nothing without the attack instruments it was tested against, who tested it and when — the reporting framework ISO/IEC 30107-3 defines.

What is being detected

Liveness detection assesses whether the face presented to a camera comes from a live person physically present. The broader term used in standards work is presentation attack detection, because what is being detected is an attack on the capture point rather than an absence of life as such — and naming it that way makes the right question obvious: an attack using what?

That question is the whole subject. Resistance is specific to the attack instruments a system has been tested against. A system that reliably rejects a photograph printed on office paper may do nothing about a high-resolution screen replay, and neither result tells you anything about a silicone mask.

There is no ISO 30107-3 certificate

ISO/IEC 30107-3 defines how presentation attack detection is tested and reported: the methodology, the terminology, and what a test report should state. It is a reporting framework, not a certification scheme, and it defines no pass mark.

So there is nothing to hold. A vendor claiming to be "ISO 30107-3 certified" is describing something that does not exist, in the same way and for the same reason as "NIST certified". What the standard does give you is the ability to compare two vendors' reports, provided both produced one.

Evaluation

What a liveness claim must state

Five things. If any is missing, the claim covers nothing you can rely on — and the last one is the one most often overlooked even by careful buyers.

What a presentation attack detection claim should state, and what its absence means.
Stated What it tells a buyer If it is absent
Attack instruments used Which specific attacks the result actually covers — printed photograph, screen replay, mask, and at what fidelity. The result covers nothing specific. It may have been tested only against the easiest possible attack.
Who performed the testing Whether it was independent, and whether the laboratory is one whose reports are taken seriously. It was probably self-assessment, which is not worthless but is not the same thing.
Date of testing Whether the result reflects the build you would actually receive. It may predate the shipped version by several releases.
Attack level or fidelity How good the attack was. A phone photograph and a high-resolution print are different tests. The result may describe resistance to an attack nobody would attempt.
Where the check runs Whether compression between the camera and the analysis has discarded the cues detection depends on. Laboratory conditions may not apply on site, and the gap can be large.

ISO/IEC 30107-3 describes a test methodology, not a pass mark. There is no certificate to hold, so a vendor claiming one is describing something that does not exist.

Threat model

Decide whether you need it before you specify it

Liveness adds cost, adds a failure mode, and can add a retry to every transaction. Those are worth paying where the threat is real and wasteful where it is not.

Unattended capture point

Usually worth it

An unattended gate or kiosk is where presenting a photograph or a screen is easiest and least likely to be noticed. This is the case liveness detection exists for.

High-value access

Usually worth it

Where the consequence of a successful substitution is severe, the cost of an extra check and an occasional retry is easy to justify.

Remote or self-service capture

Usually worth it

Nobody is watching the capture, and the person controls the device and the environment entirely. The threat is real and specifying against named instruments matters.

Capture in view of staff

Often over-specified

Someone holding a phone up to a camera in front of a staffed desk is conspicuous. The marginal value over the human already present is lower than it looks on a feature grid.

Low-consequence convenience

Often over-specified

A loyalty check-in or a staff canteen door. The cost of a failed liveness check — a retry, a queue — may exceed the cost of the attack it prevents.

Watchlist monitoring

Usually not the control

Nobody presents a photograph to a ceiling camera hoping to be recognised. The relevant controls here are quality gating and threshold policy, not liveness.

What Ayonix states

The registered claims on this page

Liveness and presentation-attack detection are available in supported capture workflows. What that means for a specific deployment is confirmed in writing rather than advertised as a rate.

ISO/IEC 30107-3 defines how presentation attack detection is tested and reported

Source type
Government or standards body
Verified
2026-09-11 · Jan Mocary, Chief Technology Officer

What this does not establish

The standard describes a test methodology. It defines no pass mark and issues no certificate, so "ISO 30107-3 certified" describes nothing that exists. Ayonix makes no compliance claim against it; it is cited as the framework a buyer should require a test report to follow.

1:1 verification, 1:N identification, watchlist matching, face tracking, and liveness and presentation-attack detection

Source type
Ayonix first-party statement
Verified
2026-09-11 · Jan Mocary, Chief Technology Officer

What this does not establish

Presentation-attack resistance is specific to the attack instruments tested for. Liveness is available in supported capture workflows, not universally.

No detection rate appears on this page

For the same reason no accuracy percentage appears anywhere on this site: a detection rate without its attack instruments, its testing laboratory and its date cannot be reproduced or compared, so it is not a fact a buyer can act on. Ask us the five questions in the table above; we would rather answer them for your deployment than publish a number that will not survive contact with your capture point.

Frequently asked questions

What is liveness detection?

Liveness detection assesses whether the face presented to a camera comes from a live person physically present, rather than from a photograph, a screen replay, a printed mask or a video. In biometric standards work the broader term is presentation attack detection, because what is being detected is an attack on the capture point rather than an absence of life as such.

What does ISO/IEC 30107-3 actually specify?

It defines how presentation attack detection is tested and reported — the methodology, the terminology and what a test report should state. It is not a certification scheme and it defines no pass mark. There is therefore no certificate to hold, and a vendor claiming to be "ISO 30107-3 certified" is describing something that does not exist. What the standard makes possible is a report you can actually compare between vendors.

What should a liveness claim state before I believe it?

Four things: which presentation attack instruments were used, at what level, who performed the testing, and on what date. A detection rate against unnamed attacks covers nothing specific — it might have been tested only against a printed photograph on office paper, which almost anything defeats. Also ask where the check runs, because compression between the camera and the analysis can remove the very cues detection depends on.

Is liveness detection always necessary?

No, and specifying it reflexively wastes money and adds failure modes. It matters most where the capture point is unattended and where the value of defeating it is high — an unattended gate, a remote verification. Where a camera is in direct view of a staffed position, a person presenting a phone screen to it is conspicuous, and the marginal value is lower. Decide from your threat model rather than from a feature list.

Does liveness detection stop deepfakes?

Those are related but distinct problems and should be specified separately. Presentation attack detection addresses something physically presented to a camera — a photograph, a screen, a mask. A synthetic video injected into the capture pipeline, bypassing the camera entirely, is an injection attack and is defended against differently, largely by securing and attesting the capture path. A vendor treating the two as one capability has not distinguished them.

Where should the liveness check run?

As close to the capture as the architecture allows. Presentation attacks are often detected through fine detail — texture, reflection, moiré patterns from a screen — and video compression between the camera and the analysis point removes exactly that detail. A liveness capability validated in a laboratory on uncompressed frames may perform quite differently on a compressed stream arriving over a network.

Does Ayonix publish presentation attack detection rates?

No, for the same reason no accuracy percentage appears anywhere on this site: a detection rate without the attack instruments, the testing laboratory and the date cannot be reproduced or compared. Liveness and presentation-attack detection are available in supported capture workflows, and the specific capability for a deployment — including what it has been tested against — is confirmed in writing rather than advertised as a number.