Facial Liveness Verification

Facial verification systems are increasingly common, from unlocking phones to verifying identity for financial services. But how do these systems know they’re looking at a real person and not a printed photo, a video replay, or a sophisticated mask?

The answer is liveness detection — a set of techniques designed to confirm that the face presented to a camera belongs to a live human who is physically present. Without liveness checks, facial verification is vulnerable to presentation attacks: an adversary could hold up a photo, play a video on a screen, or wear a 3D-printed mask to impersonate someone else.

Liveness detection generally falls into two categories:

Passive liveness analyzes properties of the captured image or video without requiring the user to do anything specific. This might include examining texture patterns (skin vs. paper), checking for micro-movements like blinking or subtle muscle contractions, or analyzing 3D depth characteristics that distinguish a real face from a flat surface.

Active liveness asks the user to perform specific actions — turn their head, blink, smile, or follow an on-screen prompt. By verifying that the face responds to randomized challenges in real time, the system can confirm that a live, cooperative person is present.

The interactive demo below implements a simplified active liveness check using randomized challenges. It runs entirely in your browser via WebAssembly — no images or data leave your device. Each session picks three challenges at random from a pool of head turns, blinks, smiles, and nods, then evaluates your responses to produce a confidence score.

Try It Yourself

Click the button below to start the liveness check. You’ll be guided through three randomized challenges using your device’s camera. The system tracks your face, measures how quickly and accurately you respond, and analyzes passive micro-expression activity to produce an overall liveness confidence score.

Liveness Detection

This liveness demo runs entirely in your browser. No images, video, or biometric data are stored or sent to any server. Refer to Privacy Policy for additional details.

How It Works

When you start the liveness check, the following happens entirely in your browser:

1. Face Detection & Tracking — A face detection model loads via WebAssembly, begins tracking facial landmarks and blendshape coefficients in real time from your camera feed, and renders an oval guide overlay.

2. Randomized Challenges — Three challenges are selected at random from a pool of five: turn left, turn right, blink, smile, and nod. Each challenge monitors the relevant signal — head pose angles for turns and nods, eye-blink blendshapes for blinks, and mouth-smile blendshapes for smiles. A challenge is passed when the signal exceeds its threshold and is held for a minimum number of frames.

3. Micro-expression Analysis — Throughout the session, the system passively samples a set of facial blendshape coefficients (eye blinks, cheek squints, brow raises) and computes their variance. A live face produces natural micro-movements; a static image or replay produces near-zero variance.

4. Confidence Scoring — After all three challenges complete, the system produces a combined confidence score weighted across three factors: challenge peak accuracy (50%), response time (20%), and passive micro-expression variance (30%). Liveness is confirmed when the confidence score meets the threshold and all individual challenges have passed.

Limitations

This is an educational demonstration, not a production-grade biometric system. It provides basic protection against printed photo attacks and static screen replays, but it is not designed to defeat sophisticated attacks like high-quality deepfake video or 3D masks. Production systems typically combine multiple signals — infrared depth sensors, texture analysis, additional challenge randomization, and server-side ML models — to achieve higher assurance levels.