Rank One is celebrating its strong performance in the National Institute of Standards and Technology’s first Face Recognition Vendor Test ranking since COVID-19 put testing on hold.
This round of testing featured 221 facial recognition algorithms, including Rank One’s ROC SDK v1.22. The algorithm showed impressive results: for example, in scanning “Mugshot” photos, its False Non-Match Rate (FNMR) was only 0.65 percent. That stat barely increased to 0.88 percent for “Visa Border” photos, and the algorithm clocked a 1.61 percent FNMR for “Border” photos.
It also showed strong results in the categories of “Template Generation Speed” and “Compare Speed”.
In fact, Rank One notes that its algorithm is in the top quintile in seven of the eight assessment categories, the exception being the category of biometric performance on “Wild” images, in which it’s ranked in the second quintile. No other vendor is ranked in the first quintile in so many categories, with the closest rivals finding their way into the top quintile across five categories.
What’s more, Rank One says that every other vendor wound up in the fourth or firth quintile in at least one category. And the company points out that some of the most high-profile players in the facial recognition market – Amazon, Clearview AI, and Microsoft – haven’t submitted their algorithms to the FRVT Ongoing evaluation for the past three years, suggesting in a blog post analyzing the results that they “should never be used for facial verification.”
Rank One first announced the SDK update in May of this year, noting at the time that it delivered a 50 percent reduction in the error rate and a 50 percent improvement in the time needed for template generation, among other improvements. That R&D has evidently paid off, given the company’s performance in the FRVT.
August 28, 2020 – The FindBiometrics Editorial Team