Liveness Detection

Liveness detection establishes that a biometric sample came from a real person physically present at capture, rather than a photograph, a mask, a replayed video, or generated media. Also called a liveness check, biometric liveness, or face liveness detection, it is what stops a stolen portrait from passing as its owner.

How it works

Two approaches, and the trade-off between them is the whole design decision.

Active liveness asks the user to do something — turn their head, blink, follow a moving target, read a phrase. The response is checked against the prompt. Because the challenge is unpredictable, a pre-recorded video cannot satisfy it, which makes active methods resistant to replay. The cost is friction and failure: users misunderstand instructions, and people with motor or visual impairments are disadvantaged by challenges that assume particular abilities.

Passive liveness analyzes a single capture without asking anything. It looks for the physical signatures of a real face in front of a real lens: skin texture at a level printers do not reproduce, subtle depth cues, specular reflection, moiré patterns and refresh artifacts that betray a screen, color distributions inconsistent with reflected light. The user experience is nothing at all, which is why it dominates high-volume consumer onboarding. The trade is that everything rests on model quality, with no challenge-response fallback.

Most production systems run passive by default and escalate to active when the passive signal is ambiguous or the transaction warrants it.

The threat has moved. Presentation attacks — holding something up to a camera — are now the easier problem. Injection attacks bypass the camera entirely, feeding synthetic frames into the capture pipeline through a virtual camera driver or an intercepted API call. No amount of image analysis detects a perfect deepfake injected as if it were a camera feed, because the image itself may be flawless. Defending against it means attesting the integrity of the capture channel rather than judging the picture.

Why it matters for identity verification

Face matching without liveness is close to worthless in a remote channel. Comparing a submitted image to a document portrait establishes that two pictures show the same person — and a fraudster holding the victim’s stolen passport can satisfy that by photographing the portrait.

Generated media has changed the economics again. Producing a convincing synthetic face of a specific person used to require skill and time; it now requires a consumer tool and a reference photo. That collapses the cost of an attack that liveness is the only control designed to stop.

The practical consequence is that liveness quality now determines the security of the whole verification flow, and it has to be evaluated on both attack families. A system with excellent presentation attack detection and no channel attestation is open to injection. Microblink’s liveness detection covers both, and injection attack detection addresses the case where the camera is never involved at all.

Active vs passive liveness

Neither family defends against injection attacks on its own. That is a property of the capture channel, not of the liveness method, and it is the most common misconception about the pair.

What it can’t do

It does not tell you who the person is. Liveness confirms a real human was present. Identity comes from the face match against an authenticated document, and the two checks fail independently — a live person can be the wrong person.

It does not stop a coerced or complicit user. Someone physically present under duress, or paid to enroll on a fraudster’s behalf, passes every liveness check that exists. Money mule recruitment relies on exactly this, and no biometric control addresses it.

Presentation attack detection does not cover injection. Image analysis judges the picture. An attack that supplies a flawless synthetic frame through a virtual camera gives it nothing to object to. Channel integrity is a separate control.

Lab certification is not field performance. iBeta Level 1 and Level 2 conformance test defined attack instruments under controlled conditions. They are meaningful and they are not a guarantee against attack types that emerge after the test was designed.

Frequently asked questions

What is the difference between liveness detection and face matching?

Face matching compares two images and asks whether they show the same person. Liveness detection asks whether the image came from a real person present at capture. Both are needed: matching without liveness accepts a photograph of the right face.

Is passive liveness less secure than active liveness?

Not inherently. Passive removes the challenge-response fallback and relies entirely on model quality, but a strong passive system outperforms a weak active one. The more useful question is how either performs against injection attacks, which neither addresses by design.

What is an injection attack?

An attack that feeds synthetic imagery directly into the verification pipeline through a virtual camera driver or intercepted API call, bypassing the physical camera. Because the image can be flawless, image analysis alone cannot detect it — defending against it requires attesting the capture channel.

Can deepfakes defeat liveness detection?

A deepfake displayed on a screen to a camera is a presentation attack and is generally detectable. The same deepfake injected directly into the capture stream is much harder, which is why injection attack detection has become the more important control.

Related reading

• Active liveness detection — challenge-response methods and their trade-offs
• Passive liveness detection — single-capture analysis with no user action
• Document liveness detection — the equivalent question asked of a document
• Liveness detection standards — what iBeta and ISO/IEC 30107 actually certify