NEC is trying to make people feel more comfortable with facial authentication technologies. To that end, the company is using homomorphic encryption to safeguard the biometric data that gets used in large-scale identity verification systems.
In concept, NEC’s solution is relatively straightforward, and is comparable to other homomorphic encryption applications. Since each user’s face data is encrypted, cybercriminals cannot use that data to generate spoofs, even if that information gets leaked in a data breach. The technology also shields user data from the service provider, since the decryption key belongs to the individual user and not to the organization performing the match. As a result, service providers cannot access or exploit the (encrypted) biometric information stored in their system, which should make users less nervous about using their face for authentication.
What sets NEC’s solution apart is the speed with which it can carry out matches with encrypted biometrics. NEC noted matching an encrypted template is more complicated than matching an unencrypted one. As a result, most homomorphic encryption systems have struggled to perform 1:N identification tasks, and have instead been limited to 1:1 matching. That makes the technology useful in online login systems where one user is tied to one account, but less appealing in bigger environments where large numbers of people need access to a certain area.
NEC’s offering dramatically reduces the processing power (and time) needed to perform a 1:N match. To do so, the system does not try to sift through every image in a database. Instead, it tries to narrow down the full database to a shortlist of likely matches. The system can generate the shortlist in roughly 0.01 seconds, and can find a match and authenticate the user in about 1 second. Those results were achieved with a database of 10,000 registered users, and when the system is able to produce a shortlist that contains about one percent of that 10,000 total.
According to NEC, the use of homomorphic encryption does not affect matching accuracy. The company is planning to integrate the technology with its Bio-IDiom authentication and access control products. In August, NEC’s facial recognition algorithm took the top spot in the NIST’s 1:N identification test with an accuracy rate of 99.78 percent.
December 16, 2021 – by Eric Weiss