2D vs. 3D Face Matching
2D Matching will never have the accuracy needed for true Unsupervised Identity Verification. There’s just too much variability in how the same 3D human face appears when flattened into 2D at different image capture distances. This variability creates significant overlapping similarity between the 2D features of different humans and confuses the 2D algorithms, preventing them from achieving highly accurate FARs at usable FRRs.
Apple, Google, and Intel understand this, so their 3D Face Matching systems use 3D infrared cameras, but that, of course, requires each device to include special hardware. In contrast, FaceTec securely performs 3D Matching from virtually any device with a 2D camera.
 |
 |
 |
|
|---|---|---|---|
| Type |
2D Software |
3D Hardware |
FaceTec® 3D SOFTWARE |
| Axes | X,Y | X,Y,Z | X,Y + Time |
| Vendors | Aware, BioID, Daon, FacePhi, Idemia, iProov, ID R&D, etc | Apple FaceID, Google Pixel 4, Intel RealSense® | FaceTec + > 90 Channel Partners Worldwide |
| Purpose | Face Matching | Unlock Mobile Phones | 3D Face Matching |
| Installed Base | 10+ Billion Smart Devices (Android-85% + iOS-14% & Webcams) | Only new iPhones have FaceID & Pixel 4 = < 12% of market | 10+ Billion Smart Devices (Android-85% + iOS-14% & Webcams) |
| Portable Biometric | Varies | None, re-enroll on each device | Cross-Device & Cross-Platform |
| Technology | Legacy 2D Matching Software | Hardware: Infrared Camera Array & Neural Network Chip | Software: Real-time Computer Vision + 100% proprietary AI |
| Interface | Varies | Glance to unlock phone | 3D Video Selfie: ~2 Seconds |
| Skin Tone Bias | Most 2D Algos have bias at published FARs | None-Reported | None observable in the Lab or Real-World usage |
| Device SDK Info | Varies | No SDK possible, special hardware required | Device SDKs for Android/iOS, web + Server SDK |
| Liveness Method | Blink, Smile, Turn Head or Flashing Lights, etc | Infrared dots + neural network chip determine if user is 3D | Measures 3D Depth, skin texture, eye reflections, etc |
| Liveness Strength | Fairly Weak | Fairly Strong | Very Strong |
| 3D Depth Detection | Weak | Very Strong | Very Strong |
| Intellectual Property | Legacy tech, too old for meaningful patents | 20+ infrared related patents acquired in 2013 | 20+ Patents on 3D process issued globally |
| FAR/FRR | Varies, but 1/<75,000 at real world usable FRRs | 1/1M - No FRR stated | 1/125,000,000 FAR @ <1% FRR |
| Identical Twin Differentiation | Very Weak | “If you have a Twin, use a PIN.” | High 1:1 FAR provides Best Possible Twin Differentiation |
| Liveness Testing Certifications | No, only non-standardized conformances, no camera feed security tested | No Official 3rd Party Testing | Certified Level 1 & 2 Spoof Detection by NIST/NVLAP LAB - Liveness.com |
| Age Estimation | 2D = poor Age Estimates | Not Available | "Better than Human" Face-only Anonymous Age Estimation |
| Match to Photo ID | Low-detail & problems with aged photos = low match rates | Not Available | Up to 1/2,000,000 Match Confidence with 3D:2D |
| Password Replacement? | Not secure enough, Liveness too Weak & FAR too low | No, only used for convenience | Yes, universal device support, highly secure & convenient |
| Spoof Bounty Programs? | No, 2D is easily spoofed | No, no motivation | $600,000, SpoofBounty.com |
