Face masks do help evade facial recognition tech—for now

Privacy newsTips & tricks
2 mins
Mask with a smiley face.

This post was originally published on August 15, 2020.

Two birds, one stone: Not only do face masks help stem the transmission of the novel coronavirus, but a new study also suggests that they might make it harder for facial-recognition software to identify individuals, too.

And when you consider that surveillance is literally everywhere in the U.S., there seems to be no reason to ever take your masks off.

[Interested in more privacy and surveillance updates? Sign up for the ExpressVPN blog newsletter.]

The study, conducted by researchers at the National Institute of Standards and Technology (NIST), a branch of the U.S. Department of Commerce, found that the failure rate of facial recognition algorithms matching people wearing masks with their unmasked photos to be as high as 50%.

The NIST conducted the study to measure the accuracy of facial-recognition algorithms submitted to it by companies and universities, as part of its Ongoing Face Recognition Vendor Test (FRVT). The technology tested was provided to the NIST before Covid-19 emerged—before anyone knew face masks would become so widely worn today.

Nine different masks, which came in black and light blue, were part of the test. Researchers superimposed the masks—some designed to cover more of the face than others—onto photos of people who were not wearing masks.

In total, the study tested 89 algorithms on over 6 million images of 1 million people.

Error rates climb

The most accurate algorithms at NIST only had a failure rate of about 0.3%—meaning they are correct 997 times out of 1,000. However, with masks present the failure rate rose to 5% for the better algorithms and reached up to 50% for the worst-performing ones. The study also notes that most algorithms had higher error rates when black masks were used as opposed to light blue masks.

The researchers add that all the algorithms evaluated are considered to be “pre-pandemic,” with the implication that algorithms might start to be made smarter to take masks into account.

In China, face scanners that can recognize people wearing face masks are already in use.

It could only be a matter of time before the technology is exported to the rest of the world, especially when you consider that over 52 governments across the globe rely on Chinese technology for their facial-recognition systems.

With or without masks in the equation, facial-recognition tech isn’t without gaping flaws. It resulted in a wrongful arrest in Michigan, and companies including Amazon and Microsoft have said they will stop selling their facial recognition technology to law enforcement, at least temporarily.

Also Read: Not so (artificially) intelligent. 8 times machine learning got it wrong

I like to think about the impact that the internet has on humanity. In my free time, I'm wolfing down pasta.