With facial masks becoming the new normal, a study has found out that the facial recognition designed to identify people with masks on are failing.
National Institute of Standards and Technology’s Findings
The US National Institute of Standards and Technology (NIST) has looked at the algorithms of 41 facial recognition applications. These algorithms were submitted around mid-March after the pandemic started raging globally. The algorithms were designed with face masks in mind and claimed to identify people with half their faces covered with masks.
NIST released a report in July which stated that face masks were preventing regular facial recognition algorithms. The error rates of these are ranging from 5% to 50%.
NIST is considered to be the leading authority on facial recognition accuracy testing. It expected algorithms to improve on identifying people with masks.
Varying Error Rates
Some algorithms had accuracy overall. For example, the Chinese facial recognition company Dahua’s algorithm error rate is going from 0.3% without masks to 6% with masks. Others had error rates increasing up to 99%.
A facial recognition provider, Rank One, had an error rate of 0.6% without masks, and a 34.5% error rate once masks were digitally applied. The company started offering Periocular Recognition in May, which claimed to be able to identify people just off their eyes and nose.
CEO of Rank One, Brendan Klare, said the company wasn’t able to submit the Periocular Recognition algorithm to NIST because of the agency’s limit to one submission per organization.
TrueFace that is used in schools and on Air Force bases saw its algorithm error rate go from 0.9% to 34.8% once masks were added. Shaun Moore, the company’s CEO, also commented on this on 12th August 2020. The company’s researchers are working on a better algorithm for detecting faces beyond masks.
Every facial recognition algorithm suffered higher error rates once masks were added. However, some error rates were as low as 3%. It indicated that it is achievable for algorithms to identify people with their faces covered.
The Future of Facial Recognition Algorithms
Face masks are proven to limit the novel coronavirus’s spread. Governments around the world have mandated people to wear masks to reduce the pandemic’s impact. Health experts are expecting the majority of people will need to continue wearing masks for years to come. This move will push facial recognition companies to improve their algorithms.
NIST is currently preparing a report on how masks have affected facial recognition algorithms. It has used 6 million images from its database and digitally added masks to the photos.
It is possible that the error rates could go higher. If NIST used real photos of people in masks, errors might range higher. The physical masks may have different shading, textures, and patterns that might confuse the algorithms further.