Unissey claims 18 percent face biometrics performance gain in latest NIST testing

To create a strong, secure and frictionless facial recognition solution, algorithms must be subjected to all kinds of attacks to test their performance level and reinforce their robustness. In a dedicated article, Biometric Update reviews Unissey’s latest results after submitting its solution for evaluation to the National Institute of Standards and Technology (USA) on November 29, 2021.

Article about UNISSEY’s evaluation to the NIST

In July 2021, Unissey, a young French start-up, submitted its facial biometric solution to the National Institute of Standards and Technology (USA) for the first time. The result? The algorithms have not been idle and were placed in the first half of the world ranking. A great success that proves the robustness and high performance of the solution for a first try!

Four months after the first evaluation, Unissey moves on to the second and finishes with an 18% performance improvement of its solution. On the VISABORDER dataset, the algorithms achieve an FNMR of 1.47% for an FMR of 10-6. Are you a little confused by all these acronyms? Don’t panic, we have prepared a small glossary to help you:

VISABORDER: The Visa-Border dataset is one of the closest scenarios to voluntary biometric authentication. 

Facial comparisons are made between photos from official identity documents, such as a passport, and border crossing images taken in real time. 

False Non Match Rate: proportion of facial comparisons between two photos of the same person that are wrongly considered as different. 

= measures the convenience of the solution: the lower this rate, the lower the risk that a person is wrongly rejected (when the comparison was made against his or her reference image, from his or her identity document for example).

False Match Rate: proportion of facial comparisons between two photos of two different people that are wrongly considered a match. 

= defines the security level of the solution: the lower the rate, the lower the risk that someone will succeed in impersonating someone else.