Stunning NIST Report: Who Knew?
July 30, 2020
Years ago I did some work for the US government. Nothing much. In the course of the work, I learned about some interesting US government reports; for example, some Library of Congress public documents which are not available to the public and a couple of studies whose subjects baffled me.
Kid Durango as a masked bank robber. Who knew that a mask would make it difficult to recognize the bad hombre?
I read “NIST Launches Investigation of Face Masks’ Effect on Face Recognition Software.” The write up reports:
Now that so many of us are covering our faces to help reduce the spread of COVID-19, how well do face recognition algorithms identify people wearing masks? The answer, according to a preliminary study by the National Institute of Standards and Technology (NIST), is with great difficulty. Even the best of the 89 commercial facial recognition algorithms tested had error rates between 5% and 50% in matching digitally applied face masks with photos of the same person without a mask.
Facial recognition works in a couple of basic ways. A system can convert a face to ovals, take calculations of the eye areas, the snoot, and the mouth. Measurements are made and the system looks for matches. As some may know, recognition accuracy can vary widely. When an artificial intelligence, super duper program is used, the oval idea is supplemented with pattern recognition. Most of the systems with which I have modest familiarity use both methods and then display possible to an investigator.
What did the NIST study conclude? Face masks reduce accuracy. Face masks cause FR systems to not recognize the face as a face. The nose plays a big part in accuracy. No nose, reduced accuracy.
The study seems well intentioned. But didn’t bank robbers in the 1840s wear masks? From my point of view, the study validates what most people already know. A person wearing a mask is harder to identify.
What’s that error rate and false positive rate again, partner?
Stephen E Arnold, July 30, 2020
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
