By Jack Morse
Well, there goes the tiniest privacy silver lining on that otherwise dark coronavirus cloud.
A brand-new research study from the National Institute of Standards and Technology (NIST) found that many commercially readily available facial-recognition systems are still able to properly determine individuals wearing face masks. As mask use becomes ever more commonplace due to the pandemic, there continues to be much police hand-wringing about possible unfavorable results on the prevalent surveillance technology. The NIST research study recommends that authorities can rest simple.
The research study, published Monday, examined the performance of 89 facial-recognition algorithms. Notably, these were all algorithms developed pre-pandemic. Simply put, none were developed for a world where masked people are the norm. However, according to NIST, some of the algorithms performed incredibly well at recognizing individuals with masks on.
” Using unmasked images, the most precise algorithms fail to verify an individual about 0.3%of the time,” discusses a news release accompanying the study. “Masked images raised even these top algorithms’ failure rate to about 5%, while numerous otherwise skilled algorithms failed in between 20%to 50%of the time.”
While that might in fact sound respectable– mask wearing increased the mistake rate, after all– the fact that it only increased it to about 5 percent for higher carrying out algorithms isn’t exactly heartening.
Thankfully, there are some noteworthy caveats to the research study which might translate to a little less doom and gloom. For beginners, the study authors– Mei Ngan, Patrick Grother, and Kayee Hanaoka– digitally included masks to photos in order to run the test. They admit that this develops a whole host of unknowns and new factors to consider.
For instance, would the differing textures of real masks impact the algorithms in some unidentified manner? Or, alternatively, how might the innovation respond to the inconsistent fit of different masks on diverse faces?
These questions are difficult to address without utilizing real images of people using real masks.
What’s more, the research study authors “did not address the impact of eye-glasses or eye-protection.” Simply put, glasses or sunglasses may provide the user with added protection against facial acknowledgment.
Do not get too comfortable, nevertheless, as the caveats go the other way, too. For beginners, remember, this study was done on pre-pandemic algorithms. NIST already has a research study in the works to look at facial-recognition innovation specifically created for the face-mask era.
” With regard to accuracy with face masks,” NIST computer researcher Mei Ngan warned in journalism release, “we expect the innovation to continue to improve.”
Likewise, there’s the possibility that– in some limited cases– masks may really make it simpler to identify topics.
” We got a tip that such info [like patterned masks] might work as a soft biometric,” keeps in mind the research study, “in that a topic that always wears the very same textured mask will be more recognizable.”
The research was performed “ in partnership with” both the Department of Homeland Security and Customs and Border Defense– a crucial reality to keep in mind as demonstrations following the killing of George Floyd continue throughout the nation, and body-armor sporting feds respond strongly to protesters in Portland, Oregon.