NIST Study Finds That Masks Defeat Most Facial Recognition Algorithms (venturebeat.com) 46
In a report published today by the National Institutes of Science and Technology (NIST), a physical sciences laboratory and non-regulatory agency of the U.S. Department of Commerce, researchers attempted to evaluate the performance of facial recognition algorithms on faces partially covered by protective masks. They report that even the best of the 89 commercial facial recognition algorithms they tested had error rates between 5% and 50% in matching digitally applied masks with photos of the same person without a mask. From a report: "With the arrival of the pandemic, we need to understand how face recognition technology deals with masked faces," Mei Ngan, a NIST computer scientist and a coauthor of the report, said in a statement. "We have begun by focusing on how an algorithm developed before the pandemic might be affected by subjects wearing face masks. Later this summer, we plan to test the accuracy of algorithms that were intentionally developed with masked faces in mind."
The study -- part of a series from NIST's Face Recognition Vendor Test (FRVT) program conducted in collaboration with the Department of Homeland Security's Science and Technology Directorate, the Office of Biometric Identity Management, and Customs and Border Protection -- explored how well each of the algorithms was able to perform "one-to-one" matching, where a photo is compared with a different photo of the same person. (NIST notes this sort of technique is often used in smartphone unlocking and passport identity verification systems.) The team applied the algorithms to a set of about 6 million photos used in previous FRVT studies, but they didn't test "one-to-many" matching, which is used to determine whether a person in a photo matches any in a database of known images. Because real-world masks differ, the researchers came up with nine mask variants to test, which included differences in shape, color, and nose coverage.
The study -- part of a series from NIST's Face Recognition Vendor Test (FRVT) program conducted in collaboration with the Department of Homeland Security's Science and Technology Directorate, the Office of Biometric Identity Management, and Customs and Border Protection -- explored how well each of the algorithms was able to perform "one-to-one" matching, where a photo is compared with a different photo of the same person. (NIST notes this sort of technique is often used in smartphone unlocking and passport identity verification systems.) The team applied the algorithms to a set of about 6 million photos used in previous FRVT studies, but they didn't test "one-to-many" matching, which is used to determine whether a person in a photo matches any in a database of known images. Because real-world masks differ, the researchers came up with nine mask variants to test, which included differences in shape, color, and nose coverage.
Sunglasses? (Score:2)
How about mask+sunglasses?
What exactly is this supposed to be used for anyway? There's no way in hell it will ever catch criminals in a world where masks are socially acceptable so that only leaves the general public.
I'm super sensitive to sunlight (Score:3)
Being a goof I got a "V for Vendetta" mask.
When I go out I look like the Unibomber's less presentable cousin.
Re: (Score:2)
Sunglasses was the first thing I thought too.
A mask is just not complete without them.
There goes Freedom (Score:2)
Re: (Score:3)
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Yes, but the research was by a deep state government department led by climate crusading scientists, so I'm sure the people that need to be convinced will still find a way to convince themselves that they are better off without masks.
Re: (Score:2)
You're right, something is broken: I keep seeing the same whiny post over and over again.
Re: (Score:2)
Back to Huffington Post with you!
Um... Duh? (Score:5, Insightful)
There's a reason why people have worn masks to hide their identities since long before computers and cameras: It works.
Re: (Score:2)
There's a reason why people have worn masks to hide their identities since long before computers and cameras: It works.
That was my thought: somebody got paid to study this? What were they expecting? There's no way these were trained with masks in mind.
Re: (Score:1)
Wouldn't you think, though, that Chinese facial-recognition software might be so trained, given the relatively longstanding social acceptance of facemasks in China?
Re: (Score:2)
When the time comes to cut the federal deficit, NIST should be on the list of agencies to trim.
If they have time for "duh" research like this, then it is obvious that they don't have enough real work to do.
Re: (Score:2)
If you can't see the value in this, I don't know what to tell you.
Re: Um... Duh? (Score:1)
Tinfoil? (Score:4, Funny)
I made my mask out of tinfoil to match my hat! I look vaguely like a disheveled knight...
Wait a minute... (Score:2)
So you're telling me that covering over 50% of your face might introduce enough error to not recognize the face?
I'd bet that wearing a full-face motorcycle helmet likely screws with facial detection too.
Re: (Score:2)
(... and police lobby governments to recognize crime-fighting is more important than keeping the population alive.)
Re: (Score:2)
I want funding for a study which shows that putting a bag over your head confuses mask recognition software.
Re: (Score:2)
It's why Daft Punk had a thriving second business of bank robbery.
Re: (Score:2)
"I'd bet that wearing a full-face motorcycle helmet likely screws with facial detection too."
Yes, but at $500 each I can't afford too many different motorcycle helmets, so the helmet might effectively identify me (or at least cut down the crowd).
COVID masks are cheap, so everyone can change who they are!
Re: (Score:2)
You fail at recognizing what is going on. I posted from Firefox on Linux, which doesn't change the fact that Slashdot still doesn't support an almost 25-year old universal text encoding standard. (Note the apostrophe in the last sentence that isn't messed up. Oh, there's another one! And another yet!)
Also, I carry an Android phone, so nice assumption except for everything about your post being completely incorrect and proving yourself to not have a single clue.
Who's game? (Score:1)
DOH! (Score:1)
So masks defeat most facial recognition programs.
Film at 11
Duh! (Score:3)
Wow, I wonder when someone came up with this. Really this took a study?
Cover half your face and it's hard to recognized you!!
Heck I have a hard time recognizing people myself...
Great News! (Score:2)
We didn't need facial recognition for 4.5 billion years. This may be the last chance we have as a society to do something about our new digital overlords knowing exactly where we are at any given time. It seems fortuitous that this mask thing happened right at the relative beginning of facial recognition.
Think its wild that the marketing folks seem to know what you're thinking and how to directly target you? Just imagine that power in people that are interested in much more than just advertising.
--
Wor
Re: (Score:2)
Actually it is obvious that we have evolved a large amount of cognition dedicated to facial recognition and also apparently there is some evolution toward more variability in faces (ie genes that significantly alter appearance while not having any detrimental physical effects). So it serves an apparently valuable evolutionary purpose. I do think it is likely to be able to recognize 40 or so "members of your group" and reliably tell if a random human was not in the group, which is somewhat different than rec
Next up (Score:2)
Opaque bags obscure their contest. Film at 11!
ICP did this (Score:2)
"Ethics" wasn't even mentioned once (Score:5, Insightful)
The aim of the research is to improve the algorithms. But wouldn't we be far better off if facial recognition algorithms were difficult, expensive and impractical?
Re: (Score:2)
The balance is always freedom vs safety. Of course, if facial recognition works then criminals, terrorists, extremists of all kinds, and bad people in general can be tracked and caught before they hurt innocents. The freedom argument is there for freedom from government tracking. But do people have privacy in public?
Just like strong encryption and gun rights, are we willing to sacrifice the (perceived, IMO) safety of prohibitions and tracking for tangible freedom?
In that respect COVID is a godsend. People w
Re: (Score:2)
That's the problem with faster-running gazelles: you get faster-running cheetahs. Good luck persuading the gazelles to not run as fast as they can, though.
Wearing the mask around your chin? (Score:2)
If you are wearing your mask below your nose, that already it narrows the ID to persons who are differently cognitively abled.
If you are wearing your mask around your chin, well, the AI shouldn't have any difficulty narrowing you down, especially if the video shows you breathing through your mouth.
This just in (Score:2)
Fear not, but level the playing field (Score:2)
This is a fight between artificial intelligence versus criminal intelligence.
We need to get criminals to believe that having a mask, i.e. just in a pocket, plays havoc with facial recognition.
We won't catch the smart criminals anyway, but there is always a chance to catch the dumb criminals.
This is very clever! (Score:1)