A terrific ‘one-in-a-million’ and an empirical ‘0.1%’, which of the two can we trust?
NIST and IARPA announced the winners of face recognition contest. The best figure for verification of 99.9% (0.1% reversely) seems to fall reasonably in the range that wouldn’t astonish anyone, although it does not look as fantastic as ‘one millionth’ that Apple boasts for Face ID.
This and the other related news that Apple’s Face ID was reportedly designed to learn to get fooled are not only eye-catching on their own but also demonstrate part of a more crucial problem.
It appears that the 'ex-factory Face IDs of low FAR with high FRR' are rapidly turning into the 'in-use Face IDs of high FAR with low FRR' day after day in a gigantic scale. Then criminals would only have to wait for a good time to come.
In any case, most critical is a fact that Face ID and other biometrics solutions are dependent on a fallback password, which only results in the level of security lower than that of a password-only authentication and also that of a biometrics-only authentication.
Biting the hand that feeds IT
