With 60-70% of your face covered up by a mask, it isn't surprising that facial recognition algorithms designed pre-pandemic aren't working very well. The U.S. government did a study that proves this as companies and security agencies scramble to design new ones.
The National Institute of Standards and Technology's (NIST) Mei Ngan says over the years the number of facial recognition software applications have exploded. "I remember one time reading about face recognition being used to ration toilet paper in public restrooms from different parts of China to combat toilet paper theft," she says.
Recently, the computer scientist conducted a study to see how reliable facial recognition algorithms were at identifying people with a mask. She applied masks digitally to millions of photographs and then compared the accuracy of masks to unmasked people. She also looked at how changing the shape, the color and the amount of nose coverage would affect the result.
"A coverage mask caused more errors when we compared them to low-coverage masks, and on average, we saw about a factor of five increase in error rates on high masks versus low masks and this was consistent across the board."
How Did Algorithms Do?
Paravision says it has developed new software that can identify people with masks and it is in the process of rolling it out.
Ngan says next will be testing algorithms that are designed for mask wearing. That is scheduled for the fall. In the meantime, she suggests businesses supplement algorithms with iris recognition or use software that focuses on the eyes, eye brows and forehead.
It appears the pandemic is certainly not slowing the use of facial recognition algorithms. The Columbus Dispatch reports Pop ID has teamed with an Ohio company, Wasserstrom, to make a program that combines instant facial recognition and temperature screening that can connect to employee's mobile devices, lock doors if a person's temperature is too high and record all that on a time log.