Having a tough time recognizing people behind a face mask? Facial recognition algorithms are finding it difficult too. As politicians and doctors are struggling to convince people to take the smallest of precautions against the virus by wearing a face mask, the department of homeland security is concerned that face masks are breaking facial recognition algorithms of the police. A study published by an agency showed that even the best facial recognition systems have an error rate of 5% to 50% when identifying masked faces.
The continuous threat of the coronavirus and its rapid spread around the globe has created a barrier in facial recognition’s global expansion as everyone is covering their faces. Even before the pandemic in so-called ideal conditions facial recognition technologies struggled with their accuracy and have a miserable record when identifying faces that are other than male or white. It can be said that the pandemic has given the police a brand-new crisis.
This mask problem is the reason that Apple has made it easier by enabling its users to unlock their iPhones without using Face ID. The algorithms that facial companies use rely significantly on facial markers around the nose, mouth, and eyes. However, while wearing face masks the mouth, nose, and majority of the chin and cheek are covered which makes it difficult to identify any person. The companies are now trying to upgrade their algorithms so that they can solely focus on identifiable markers around the eyes.
The study conducted by the US National Institute of Standards and Technology (NIST) examined the algorithms which were created before the coronavirus pandemic and now the researchers are assessing ways in which the effectiveness of the algorithms in detecting masked individuals can be improved.
NIST conducted a test of various software by drawing masks on the photos of people from a trove of border crossing photos. These were uploaded into facial recognition technologies together with unmasked photos from the same source. The agency scanned nearly 6 million images of 1 million people by making use of around 89 algorithms supplied by academic labs and tech firms.
NIST’s study shows that if the conditions are ideal the rate of failure is around 0.3 percent for the best recognition systems. However, this figure varies on the basis of a person’s race, age, and gender, and with a mask the failure rate increases to 5%.
Even before this pandemic governments of some countries were pursuing the technology that is capable of recognizing people trying to conceal their faces behind a mask. Face masks become a trademark for the protestors in Hong Kong because of fears of being arrested if they were identified and also to protect themselves against tear gas. Due to this the government of Hong Kong imposed a ban last year on face masks at all the public gatherings and if the protestors refused to remove the mask on police’s orders, they would have faced a potential jail term of 6 months.
Privacy Activists are looking for creative ways to conceal their faces. In London, artists who were not in favor of high surveillance, in an attempt to fail face detection systems, painted their faces with geometric shapes. Then as the covid19 outbreak came the health experts around the globe started encouraging people to wear masks covering nose and mouse.
NIST’s study shows that the way in which people wear masks and what masks they wear has a big impact on facial recognition systems. The results are unsurprising as the more facial features covered the harder it is for the system to recognize.