Liveness Detection

Proving You're
Actually There

Multi-modal liveness verification that defeats photos, videos, deepfakes, and 3D masks. Our passive + active challenge system ensures only real, present humans can verify their identity.

100%

Spoof Detection

< 3s

Check Duration

False Positives

LIVENESS CHECKStep 3/8

Turn your head right

Follow the instruction slowly

Blink Detected

Head Turn Left: Passed

Head Turn Right: In Progress

Smile Detection: Pending

Anti-Spoof Active
Texture + depth analysis running

Challenge-Response Protocol

Active Liveness Challenges

Randomized multi-factor challenges that are impossible to pre-record or fake.

Blink Detection

Eye Aspect Ratio (EAR) analysis detects natural blinks. The system measures the precise timing and symmetry of eyelid closure — something photos and static deepfakes cannot replicate.

Method: 68-point facial landmark tracking with EAR threshold at 0.21. Requires 2+ consecutive frames below threshold.

Head Pose Estimation

Users are prompted to turn their head in a random direction (left, right, up, or down). The system tracks 3D head rotation angles to verify real spatial movement.

Method: PnP (Perspective-n-Point) algorithm on 6 anchor landmarks. Yaw/pitch thresholds: ±15° to validate genuine motion.

Expression Analysis

Random micro-expression prompts (smile, raise eyebrows) validate fine-grained facial muscle control. Pre-recorded videos are defeated because the specific expression is randomized per session.

Method: Lip corner ratio + nasolabial fold depth analysis. 4 randomized expression prompts per verification.

Passive Analysis

Invisible Anti-Spoof Layers

Running silently in the background — the user never sees these checks.

Texture Analysis

LBP (Local Binary Patterns) on facial skin
Detects paper/screen pixel grid artifacts
Micro-texture comparison vs. real skin
Moiré pattern detection for screen replays
Operates on raw camera frames (no filters)

Lighting Forensics

Specular highlight analysis on skin
Light source direction consistency
Screen backlight glow detection
Flash reflection analysis (if available)
Ambient light sensor correlation

Temporal Coherence

Frame-to-frame motion consistency
Natural micro-movement detection
Video replay FPS anomaly detection
Compression artifact analysis
Motion parallax depth estimation

Depth Estimation

Monocular depth from facial geometry
Inter-pupillary distance validation
Nose bridge protrusion analysis
Face curvature vs. flat surface detection
Edge sharpness at face boundaries

Threat Model

Attacks We Defeat

Every known presentation attack vector has been tested and neutralized.

Attack Vector	Method	Detection Layer	Status
Printed Photo	Holding a printed photo of the victim	Texture analysis + blink detection	✓ Blocked
Screen Replay	Displaying victim's photo on a screen	Moiré detection + lighting forensics	✓ Blocked
Video Replay	Playing a video of the victim	Temporal coherence + random challenge	✓ Blocked
Deepfake (Real-time)	AI-generated face overlay in real time	Compression artifacts + micro-expression timing	✓ Blocked
3D Mask	Silicone or 3D-printed face mask	Texture analysis + specular highlights + blink	✓ Blocked
Cutout Mask	Photo with eye/mouth holes cut out	Edge discontinuity + depth estimation	✓ Blocked
Twin/Lookalike	Identical twin attempting verification	ArcFace embedding distance threshold	✓ Blocked
Camera Injection	Virtual camera feeding pre-recorded frames	Device attestation (TEE) + camera fingerprint	✓ Blocked

Technical Superiority

Why Our Liveness is Unmatched

Multi-layered defense that no single attack vector can penetrate.

PollGuard Liveness

Passive + Active dual-layer defense

Randomized challenge sequences (unpredictable)

On-device processing (no network dependency)

TEE device attestation blocks virtual cameras

Texture + depth + temporal + lighting analysis

Completes in under 3 seconds

Works on any Android 8+ / iOS 14+ device

ISO 30107-3 PAD Level 2 aligned

Typical Industry Approach

Single-layer check (usually just blink)

Fixed challenge order (easily scripted)

Server-dependent (fails offline)

No virtual camera protection

Only texture OR depth (not both)

8-15 second processing with retries

Requires high-end camera hardware

Self-certified without ISO alignment

Performance

Measured Results

Tested against 340,000+ live verifications and 50,000+ simulated attacks.

100%

Spoof Block Rate

All attack vectors neutralized

99.2%

Real-User Pass Rate

Genuine users pass first attempt

2.4s

Average Duration

Camera open → liveness confirmed

0.0%

False Accept Rate

Zero spoofs accepted in production

Proving You'reActually There