Skip to content

3D Mask Attacks

Definition

3D mask attacks use physical three-dimensional replicas of a victim's face to defeat liveness detection. These are the most sophisticated presentation attacks, ranging from cheap paper-craft to expensive custom silicone.


Mask Types

Type Material Cost Realism Detection Difficulty
Paper-craft Folded/curved paper $5-20 Low Easy-Medium
3D-printed (rigid) PLA/resin $50-200 Medium Medium
Resin cast Polyurethane resin $100-500 Medium-High Hard
Silicone (custom) Medical-grade silicone $300-3,000 Very High Very Hard
Transparent overlay Thin silicone over real face $200-1,000 High Very Hard

Detection Approaches

Method What It Detects Hardware Needed
Material analysis Skin vs silicone/resin reflectance RGB camera
Thermal imaging Masks don't emit body heat Thermal camera
NIR sub-surface scattering Light passes through skin differently than synthetic materials NIR sensor
Micro-expression analysis Masks limit expression range and speed RGB camera + temporal analysis
Eye movement tracking Rigid masks restrict natural eye movement High-fps camera
Depth sensing Detect mask edges, seams, unnatural depth Structured light / ToF sensor

Datasets for 3D Mask Research

Dataset Mask Types Key Feature
HiFiMask Resin, plaster, transparent Most realistic, 54K videos
WMCA Rigid, flexible, paper-craft Multi-spectral (RGB+Depth+NIR+Thermal)
3DMAD Paper-craft 3D masks Early 3D mask dataset
SiW-M Various mask categories Part of larger multi-attack dataset

Key Takeaways

Summary

  • 3D masks are rare but dangerous — silicone masks can defeat many RGB-only liveness systems
  • Multi-spectral sensors (NIR, thermal, depth) significantly improve mask detection
  • iBeta Level 2 specifically tests against 3D mask attacks
  • HiFiMask dataset provides the most realistic mask attack data for research