Facial recognition technology has started catching on in the past few years. However, the COVID-19 pandemic has put its emergence on hold. The problem isn’t the virus itself. It isn’t even because people are going out in public less often.
Instead, the face masks being used to prevent the spread of COVID-19 render facial recognition useless. Without being able to see someone’s facial features, the tech can’t accurately identify them.
Depending on who you ask, that is either a great thing or highly problematic. Given the fact that facial recognition is increasingly being used for law enforcement and security purposes, researchers have been working tirelessly to improve how it works when face masks are involved.
Results from a new U.S. government study show that those efforts are paying off.
Identification Around the Mask
Masks pose a serious challenge for facial recognition. In a normal world, it wouldn’t be a problem since only a tiny fraction of people wear masks on a regular basis. In the midst of a global pandemic, it is a major obstacle.
Although face masks likely won’t be around forever, experts have been working to develop solutions that allow facial recognition algorithms to work in the meantime.
The accuracy rates of facial recognition technology are monitored by the U.S. National Institute of Standards and Technology (NIST). Since May 1, the agency has been conducting a series of studies to determine how face masks affect the tech.
Earlier this year, it found that masks increase facial recognition error rates by up to 99 percent—even for algorithms designed to operate around face coverings.
On Tuesday, the agency published results that suggest facial recognition is getting better at co-existing with masks. The study considered 65 unique algorithms that were tested with a combined 6.2 million photos.
Without masks, the algorithms had an error rate of 0.3 percent. However, once masks were added to the photos, the rate jumped to five percent.
That is a noteworthy improvement over pre-pandemic algorithms. Researchers note, “The current performance of face recognition with face masks is comparable to the state-of-the-art on unmasked images in mid-2017.”
The newfound success can be attributed to algorithms that are designed specifically for mask wearers. Those in the NIST study look at things like the position, shape, and the distance between a person’s eyes.
More Work to Do
Despite the promising results, facial recognition is still far from accurate in the real-world when masks are involved. Poor photo quality, unpredictable lighting, and varying angles can all disrupt algorithms that are barely able to identify masked faces in a controlled setting.
Moreover, the NIST study didn’t account for differences between people of various races and gender. The agency said, “We deferred tabulating accuracy for different demographic groups until more capable mask-enabled algorithms have been submitted to [the Facial Recognition Vendor Test].”
That will be an important thing to consider in the coming days since facial recognition technology is notorious for inaccurately identifying women and people of color.
Although the mask-related setback is bad for companies that are trying to push facial recognition, many people aren’t upset. In fact, some see wearing a mask as an added bit of privacy.
For those individuals, it’s worth noting that certain colors of masks make it harder to identify the wearer. The NIST study found that red and black masks do a better job of tricking facial recognition than blue and white.
It will be interesting to see how facial recognition companies continue responding to the challenges posed by masks. Indeed, the pandemic may be over before a viable solution is found.