Hayes via NIST)
Facial-recognition technology is controversial at best.
Add pandemic-era masks and the algorithms are easily flummoxed, according to a study from the National Institute of Standards and Technology (NIST).
Of the 89 systems tested, even the best ones failed to match digitally applied masks to a control portrait of the same uncovered face as often as half of the time, the report finds.
"With the arrival of the pandemic, we need to understand how face recognition technology deals with masked faces," says Mei Ngan, co-author and NIST computer scientist.
"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 research team explored each algorithm's capacity for "one-to-one" matching—comparing two photos of the same person, as with smartphone unlocking or passport checks—by digitally applying various mask shapes to the original image.
Of course, real-world masks differ (as does each person's understanding of how to wear them): Most folks sport surgical or reusable cloth masks, while some simply tie a bandana around their face.
Others, meanwhile, have opted for something a bit more professional.
NIST implemented nine variants, including differences in shape, color (light blue or black), and nose coverage.
"We can draw a few broad conclusions from the results, but there are caveats," Ngan explains.
"None of these algorithms were designed to handle face masks, and the masks we used are digital creations, not the real thing." Keeping that in mind, the study suggests overall accuracy with masked faces has declined "substantially."
Using unmasked images, algorithmic failure rates are about 0.3 percent; cover someone's nose and mouth, though, and even the top systems strike out 5 percent of the time.
That number reaches 20 to 50 percent for "otherwise competent" programs.
Researchers also found that dark, full-coverage masks tend to confuse algorithms the most, blocking the system from completing its usual process of measuring and comparing a face's features.
"With respect to accuracy with face masks, we expect the technology to continue to improve," Ngan says.
"But the data we've taken so far underscores one of the ideas common to previous [face recognition vendor] tests: individual algorithms perform differently." Moving forward, the team plans to test programs that take face masks into account; future studies will include "one-to-many" searches and other variants designed to broaden results.
