Multi-Spectral Liveness¶
Definition¶
Multi-spectral liveness uses sensors beyond standard RGB cameras â including Near-Infrared (NIR), depth (ToF/structured light), and thermal â to detect presentation attacks based on physical properties invisible to RGB cameras.
Sensor Types¶
| Sensor | What It Captures | Spoof Detection Advantage |
|---|---|---|
| NIR (Near-Infrared) | 850-940nm reflected light | Skin has unique NIR reflectance; paper/screen differ dramatically |
| Structured Light | 3D depth via projected dot pattern | Detects flatness of 2D attacks, mask edges |
| Time-of-Flight (ToF) | Distance measurement per pixel | 3D shape verification |
| Thermal | Heat emission (8-14Ξm) | Real faces emit heat; masks/prints don't |
| SWIR | Short-wave infrared (1-2.5Ξm) | Deep material classification |
Device Availability¶
| Sensor | Consumer Phones | Dedicated Hardware |
|---|---|---|
| NIR | Limited (some phones have IR) | Widely available |
| Structured light | iPhone FaceID, some Android | Yes |
| ToF | Samsung, some flagships | Yes |
| Thermal | FLIR add-ons only | Specialized devices |
Key Takeaways¶
Summary
- Multi-spectral approaches are significantly more robust than RGB-only, especially against 3D masks
- NIR is the most practical additional sensor â detects material differences RGB cannot
- iPhone FaceID (structured light + NIR) is the gold standard for consumer multi-spectral liveness
- Most eKYC operates RGB-only because users have diverse phone models â multi-spectral is for premium/dedicated devices
- For RGB-only systems, strong AI models must compensate for the lack of additional sensor data