How active liveness detection works

A photo of a face and a photo of an ID are not enough to know a real person is sitting in front of the camera. Someone could hold up a printed photo, replay a video on a second screen, or wear a mask. Liveness detection is the step that answers a narrower, harder question: is there a live human present right now, not a recording or a replica of one? This post explains how active liveness works, how it differs from passive liveness, and what it's actually defending against.

Active vs. passive liveness

Liveness detection generally comes in two flavours. Passive liveness analyzes a single image or short clip for signs it isn't a live capture — texture, reflections, moiré patterns from a screen, depth cues — without asking the person to do anything. It's fast and low-friction, but it's also working from less evidence: one frame, or a few, to make the call.

Active liveness asks the person to do something a static image or a pre-recorded video cannot convincingly fake on demand: blink, turn their head and return to centre, nod, or smile, on cue, during a live session. ShieldID uses active, challenge-based liveness detection, tracked across at least ten frames using 478-point 3D facial landmark tracking — not a single photo check. The system is watching how the face moves through the challenge, not just what it looks like in one still image.

Why a challenge instead of one selfie

A single selfie only has to fool the system once, in one frame. A guided challenge has to be fooled continuously, across a sequence the system chose at random, tracked on hundreds of points on the face as it moves. A printed photo does not blink or turn. A replayed video is locked to whatever motion was originally recorded — it can't respond to a challenge issued after the fact. A static mask does not move the way skin and muscle do across a nod or a smile. None of this makes active liveness unbeatable — no single check is — which is exactly why it runs alongside separate checks, not instead of them.

What runs alongside the liveness challenge

Liveness is one signal among several that run on every selfie, every time — never as a paid upgrade or a toggle a business can switch off to save money. Anti-spoofing, deepfake detection and 3D-mask detection each run as their own, separately reported check, rather than being folded into a single opaque liveness score. That matters in practice: if a verification fails, you can see whether it was the liveness challenge, a spoof signal, a suspected deepfake, or a mask detection that flagged it, instead of a single “failed” with no explanation.

• Anti-spoof: signals associated with presenting a non-live source to the camera — a screen, a print, a mask — checked independently of the liveness challenge itself.
• Deepfake detection: signals associated with a synthetically generated or manipulated face, run as its own check.
• 3D-mask / print detection: checks aimed specifically at physical presentation attacks — a printed photo or a mask held up to the camera.
• Face match: a separate step that compares the live capture against the document portrait, reported against a fixed decision line rather than a raw percentage.

What a presentation attack looks like

In the language of the field, an attempt to fool a biometric system by presenting something other than a live, cooperative person — a photo, a video replay, a mask, a deepfake — is called a presentation attack. Active liveness, anti-spoof, deepfake and mask detection each defend against a different flavour of presentation attack: a static challenge defeats a still photo; frame-by-frame tracking makes a pre-recorded replay harder to pass because it can't react to a challenge chosen live; mask and deepfake detection target the more sophisticated attempts that try to move convincingly.

Being honest about the limits

We do not publish an accuracy percentage for liveness detection, and we are not claiming a formal certification like iBeta or ISO/IEC 30107-3 — we don't hold one today, and we won't say we do. What we can say precisely is what runs and how: an active, multi-frame challenge tracked on hundreds of facial landmarks, backed by separate anti-spoof, deepfake and mask checks on every single verification, with the result reported as its own signal rather than folded into one score you can't inspect. If and when formal testing exists, we'll publish the result and the date, not a badge with no evidence behind it.