Paravision’s facial recognition algorithm is continuing to perform well in recent rounds of NIST testing. The company’s error rate went down an additional 25-30 percent in the latest 1:N test, which was enough to allow Paravision to retain its standing with the most accurate facial recognition solution from a vendor based in the US, the UK, or the European Union.
According to Paravision, the results are particularly noteworthy because the version of the platform that the company submitted was built without using any data from the now-defunct Ever photo storage platform. The company shut down the photo sharing service in August, and released its fourth-generation facial recognition solution roughly one month later.
Digging into the numbers, Paravision was the top-ranked Western provider in a number of different categories, with a performance that was two-to-three times better than its US, UK, and EU competitors in many cases. That remained true when matching mugshot, webcam, and border photos, and when using large databases that contained upwards of 6 million images. Paravision also did well when asked to account for an age gap, and when performing matches under challenging conditions, as is the case with a profile image taken at a sharp angle.
Paravision achieved a top-five overall score in each of those categories, with the exception of the border category, where it placed sixth. The company posted similarly strong results during the NIST’s previous 1:N test, which was completed in March of 2020, and in 1:1 testing conducted in 2019. Paravision has now received top-five overall scores on both evaluations.
“Achieving this level of accuracy isn’t just about delivering safety and security,” said Paravision CPO Joey Pritikin. “It’s also about enabling our partners to deliver transformative customer experiences and outstanding efficiency across the widest range of users and use cases.”
In other news, Paravision has established a set of ethical AI Principles and appointed a Chief AI Ethics Advisor to guide the development of its facial recognition technology moving forward.
March 4, 2021 – by Eric Weiss