Face recognition algorithms work by measuring a face’s features – their size and distance from one another, for example – then comparing these measurements to those from a photo stored in a passenger’s ePassport or travel document. As this technology began gaining traction at airports, COVID-19 threw a wrench in the works, making masks mandatory during air travel. Vendors including Thales, NEC and EDGENeural.ai, are adapting their algorithms to improve facial accuracy for masked subjects.
Concerted efforts to introduce biometric technologies to screen passengers as they traverse airport checkpoints have become ubiquitous in the wake of the terrorist attacks of September 11, 2001.
“It was the 9/11 Commission and subsequent legislation that called for the implementation of biometrics in arrival and departure operations to and from the United States,” Kimberly Weissman, senior communication advisor at the U.S. Department of Homeland Security, Customs and Border Protection (CBP), told APEX Media. “To date, CBP has processed more than 52 million travelers and captured almost 300 imposters through facial biometrics,” Weissman said, adding that a new website has been launched “to inform the public and our key stakeholders about CBP’s mission to expand facial biometrics to further secure and streamline touchless travel.”
In July, the National Institute for Standards and Technology (NIST) released a report on face recognition accuracy with masks following a series of trials using pre-COVID-19 algorithms, to gauge the extent to which existing biometric systems cope when faces are partially obscured. This was the first in a planned series from NIST’s Face Recognition Vendor Test (FRVT) program, conducted in collaboration with the Department of Homeland Security’s Science and Technology Directorate, Office of Biometric Identity Management and CBP. “Results were not as accurate with masks on, and there will be additional testing now that the [facial recognition] vendors are training their algorithms with masks in mind,” Weissman explained.
NIST tested 89 commercial facial recognition algorithms on a set of 6 million photos. The best-performing algorithm had error rates between 5% and 50% in matching digitally applied face masks with photos of the same person without a mask. “With the arrival of the pandemic, we need to understand how face recognition technology deals with masked faces,” said Mei Ngan, a NIST computer scientist and one of the report’s authors. “Later this summer, we plan to test the accuracy of algorithms that were intentionally developed with masked faces in mind,” she said.
In May, NEC announced the development of a touchless, multimodal biometric authentication terminal which it said achieved “the world’s highest level of precision,” during NIST testing. It uses a combination of face and iris recognition technologies from NEC’s Bio-IDiom system, which enabled authentication with a false accept rate of less than one in ten billion during the tests.
To reliably and accurately capture iris information, which is harder to obtain in comparison to face information, NEC developed a technology that locates the position of the iris based on face information, then automatically adjusts the focus and lighting, enabling authentication in approximately 2 seconds. NEC said that its technology enables identification of individuals “even when gloves and face masks are worn.”
“We can obtain a recognition accuracy of mask wearers of around 99%.” – Philippe Faure, Thales
At Thales, two types of tests have been carried out using its Thales Cogent FRP (Face Recognition Platform). One is to see if it is able to detect whether a person is wearing a mask or not, the other is to determine if it is able to recognize a person if he or she is wearing a surgical mask (The technology is 99.9% accurate when no mask is present.)
“For the first scenario, we have been able to train our algorithms so that they can detect whether people are wearing – or not wearing – a surgical mask, with a success rate above 96%,” Philippe Faure, business owner of Thales Face Recognition Platform, told APEX Media. “In the second scenario, we can obtain a recognition accuracy of mask wearers of around 99% as long as the reference photo that we use is of good quality, such as in a passport or an ID document, and provided the mask is worn in a manner that does not hide the eyes.” To get to that level of precision, Thales had to do some additional training of its deep neural network algorithms with people wearing masks so that the system focuses more on the eye area and the parts above the mask to attempt recognition.
In Pune, India, ‘deep-tech’ startup EDGENeural.ai told APEX Media that its FacEDGE facial recognition technology can work with and without masks. “We believe use of masks will continue for a prolonged period in any public place around the world as a preventive measure, not just limited to the current pandemic but also any future infections,” Sarvesh Devi, the company’s co-founder said. Having modified its “deep-learning-based facial recognition algorithm and hardware,” Devi added that the “patent-pending mask-based facial recognition technology is up to 95% accurate and reliable.”