FaceTec’s 3D face biometrics algorithm called “the best performing 1:1 face matching algorithm in history.”
Biometric face authentication specialist FaceTec announced a major benchmarking breakthrough today, issuing a press release stating its ZoOm 3D FaceMap technology is is 668-percent better performing than the operating point of the National Institute of Standards and Technology (NIST) #1 Leaderboard entry. Boasting a real-world false acceptance rate (FAR) of 1/4,200,000 with less than a one-percent false rejection rate (FRR), the ZoOm algorithm is staking its claim as “the best performing 1:1 face matching algorithm in history.”
“When we compared [ZoOm algorithms’] performance, the operating point (combination of FAR & FRR) for our 3D face matching algorithm was at least 668% better than the operating point reported by the #1 NIST algorithm,” said Josh Rose, CTO of FaceTec.
Biting the Apple
To put the numbers into perspective, Apple’s Face ID 3D biometric authentication, which uses specialized VCSEL hardware to map a user’s face, claimed to have an FAR of 1/1,000,000 when it was launched. And while Apple, as always, remains tight lipped about exact numbers and methodology as it quietly releases updates to its biometric security, FaceTec’s newly released stats are over four times as accurate as the original Face ID.
The key differentiator between Face ID and ZoOm, however, is that Apple’s solution is hardware dependent, while FaceTec’s is a software that can be installed on any device with a camera. Rather than using specialized sensors, ZoOm creates a 3D map out of two-dimensional video by asking a user to simply move slightly closer to a smartphone camera or webcam.
“ZoOm’s 3D FaceMaps are created with standard 2D cameras, but contain much more data than a flattened photo ever can,” said Kevin Alan Tussy, CEO of FaceTec. “Think of copper phone lines used for both dial-up and DSL. Just as the latter achieves vastly better results, ZoOm utilizes the same ubiquitous camera hardware to outperform 2D face matching in every way.”
Testing Needs to Keep Up
FaceTec’s benchmarking is a feat in and of itself. The NIST ongoing Face Recognition Vendor Test (FRVT), which has been underway in some form for fifteen years, relies on 2D face datasets for performance testing. That means any 3D face authentication technology – not just ZoOm – can’t properly undergo the FRVT and expect any meaningful results. FaceTec took matters into its own hands, creating a NIST-like set containing 3D face images. The results and testing methodology are laid out in ZoOm® 3D Face Matching Self-Certification Report, which also contains an illuminating FAQ about the significance of the results.
ZoOm’s self-certification comes after FaceTec’s series of third party liveness detection certifications from iBeta, and an intense campaign of education on the importance of accurate benchmarking and transparent industry standards. The fact that FaceTec had to rely on self-assessment for FRVT comparable FAR and FRR benchmarking highlights a major need for the biometrics industry: as authentication evolves to rely on new imaging, matching and presentation attack detection models, testing initiatives need to adapt to best adjudicate their accuracy.
June 19, 2019 – by Peter B. Counter