Study proves masked faces make recognition technology less accurate

An initial study by the National Institute of Standards and Technology (NIST) in the US has shown an error rate as high as 50% in facial recognition for people wearing face masks. 

The government body tested 89 algorithms and found them to be between 5% and 50% inaccurate when comparing a photo of someone with a digitally applied face mask and a picture of that person unmasked.

The most accurate algorithms in use at border crossing fail to authenticate about 0.3% of unmasked people, the NIST said in its findings.

The study’s research was carried out before the coronavirus pandemic made wearing face masks commonplace.

The just-released report is planned as the first in a series assessing different aspects of masks on the performance of facial recognition technology. 

“Given algorithm-specific variation, it is incumbent upon the system owner to know their algorithm. While publicly available test data from NIST and elsewhere can inform owners, it will usually be informative to specifically measure accuracy of the operational algorithm on the operational image data collected with actual masks,” the report’s authors wrote.

 


Registration now OPEN for PrivSec Global
Taking place across four days from 30 Nov to 3 Dec, PrivSec Global, will be the largest data protection, privacy and security event of 2020.

Reserve your place today and gain access to the entire event free of charge. With all sessions available to view live or on-demand, you can build a personalised agenda based on your key focus topics and make the event fit around your work schedule.

We have been awarded the number 1 GDPR Blog in 2019 by Feedspot.