Paravision, the US-based leader in mission-critical computer vision, announced today that it achieved top 2 accuracy globally on the National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT) July 27 report focused on face recognition with masks. With error rates below 3% for faces where the nose and mouth are covered, Paravision performed within 0.5% of the top-ranked provider, and was the #1 provider from the US, UK, Europe, and Japan by a substantial margin.
The latest report from NIST, titled “NISTIR 8311 – Ongoing Face Recognition Vendor Test (FRVT) Part 6A: Face recognition accuracy with masks using pre-COVID-19 algorithms,” builds upon NIST’s gold standard tests for face recognition, which have focused on a range of topics, including 1:1 verification, 1:N identification, and demographic effects on performance. Based on the rapid emergence of COVID-19 and the associated broad use of face masks for public health, the performance of face recognition in this scenario has quickly become critical to assess. To respond, NIST used existing datasets and synthetically added face masks of different colors and levels of coverage.
Paravision’s AI-based face recognition algorithm performed exceptionally well across all mask types. The test looked at full-width face masks in light blue and black as well as round masks in light blue. When worn slightly below the nose, Paravision demonstrated False Non-Match Rates (FNMR) of 1.24% to 1.35% at a False Match Rate (FMR) of 0.001% (1 in 100,000). When worn covering the nose, Paravision demonstrated FNMRs of 1.81% to 3.27% at an FMR of 0.001%. For light blue, full-width face masks worn specifically above the nose, Paravision demonstrated an FNMR of 2.81% at an FMR of 0.001%.
“We’re very proud to be representing AI leadership in the US,” said Doug Aley, CEO of Paravision. “Our team is dedicated to using best-of-breed technology and processes, harnessing the best of what deep learning has to offer, while continually challenging ourselves to tackle the toughest problems in computer vision. Rather than needing to re-engineer our technology to work well with face masks, our consistent focus on accuracy and robust performance has positioned us to deliver high-performance, touchless identity on an immediate basis.”
The strong results build upon Paravision’s success with NIST FRVT results in other categories. Paravision is ranked #3 globally in 1:N identification, and is the #1 provider from the US, UK, and Europe on both 1:1 verification and 1:N identification. Paravision has also shown leadership in other extreme tests, delivering #1 ranked accuracy in profile (90 degree) matching, with error rates at approximately half of the second closest in the world.
“So many aspects of NIST FRVT provide value and insight to Paravision and the rest of the industry,” said Joey Pritikin, Chief Product Officer at Paravision. “While we’re always thrilled to perform well, evaluations that assess performance in the toughest situations—like heavy angles or with face masks—are the most exciting to us, as they support what we consistently find in the field: We shine in challenging, real-world conditions.”
High performance face recognition with masks is one of several computer vision solutions Paravision provides for a world challenged by COVID-19. In addition, Paravision recently delivered a face mask detection toolset which can be used to ensure proper mask usage across a wide range of mask types. Combined, face recognition and mask detection can enable touchless safety, security, and identity in travel, workplace access, retail, and beyond.