When the task of matching a photo of a person wearing a digitally paired face mask is done According to the report, a different picture of a person without one, the most accurate facial recognition algorithm, failed to make a correct match between 5% and 50% of the time. Generally being said that most of the algorithms tested had failure rates of between 20% and 50%, Mei Ngan, a computer scientist at NIST and an author of the report, told CNN Business.
Identification issues are understandable, as facial recognition systems typically work by comparing measurements between different facial features in one image to those in another. Closing a part of the face means that there is less information for the software to match.
For their report, the researchers created nine different Black and light blue masks are shaped in ways that mask shapes differ from those in the real world and are used to hide part of a person’s face in a photo. They then digitally compared masked photos of each person with each other, a masked photo of the same person. They also tested the algorithm on both sets of photos with no virtual mask.
In total, they tested 89 algorithms on more than 6 million photographs, with a million different ones. The photos came from two sources: US applications for immigration benefits, which were used as unpublished images, and photographs of travelers crossing the border to enter the US, who were given a digital mask.
NIST found that the best of these algorithms — which was presented in the lab before mid-March — failed only 0.3% at the time of testing on these same sets of photos without a digital mask. However, with digital masks, the error rate between these same algorithms reached 5%.
One obvious drawback of the report is that NIST did not test the algorithm on images of people who were actually wearing masks – Ngan said the digital approximation of the masks was used because of time and resource constraints. On the plus side, this allowed researchers to gain an understanding of the effect of masks on algorithms, but real masks fit differently on different people. It is still unknown how textures or patterns can affect the accuracy of facial recognition software.
“That’s something we want to see,” Ngan said.
It is the focus for Marios Savvidis, a professor at Karyogi Mellon University, who studies biometric identification. He added that the person wearing the mask may be essentially invisible to the facial recognition system, as it does not detect the face at first. He thinks that the area of the face that includes the eyes and eyebrows changes little over time, which makes up a good part of the face when trying to identify someone whose mouth and nose are hidden Occur.
The NIST report is the first of many that the lab plans to release on how to identify facial masked facial masking algorithms. In the fall, Engan said, NIST hopes to issue a report on the accuracy of algorithms that were created specifically with the intention of spotting people in masks.