Facial recognition firms are scrambling to see around face masks

With the new corona virus, face detection providers expected to install their technology anywhere: at airports, casinos, restaurants, and schools. Face masks threaten to change all of this, but industry sees the situation as a speed limit rather than a roadblock.

Some companies claim that their technology is not affected by masks, and it is artificial intelligence can still recognize and identify people with a high accuracy rate even when half of the face is covered.

A public beta program for Apple’s latest iOS version showed that the technology giant is updating its facial ID to accommodate people with masks. Google did not respond to a request for a comment as to whether this is also possible with Android.

Facial recognition experts are skeptical of claims that the technology is not affected by masks. After all, face detection can trip even without a mask – studies have shown that the majority of face detection algorithms had a 10 to 100 times higher rate of false alarms for colored people.

Because of the pandemic, these algorithms cannot be properly tested with face masks by the U.S. National Institute of Standards and Technology (NIST), which many consider to be the leading authority on face recognition accuracy.

However, face detection is proposed as a solution for COVID-19 without having to demonstrate that the surveillance measure has benefits or even works properly with masks on.

“These workarounds are part of a greater effort to make an ever-expanding surveillance infrastructure a fundamental part of COVID-19 response control,” said Evan Selinger, professor of philosophy at the Rochester Institute of Technology, in a statement.

Identifiable?

Masks have long been a method to avoid facial recognition. The protesters in Hong Kong relied on them to surpass government facial recognition, which resulted in one Mask ban there.

“The largest amount of biometric data that makes us unique is in the central part of the face, just above the forehead line to the chin,” said Eric Hess, senior director of facial recognition product management at SAFR. “When we put face masks on, we block access to a significant amount of data points that help us distinguish one person from another.”

With face masks common today, several facial recognition companies have indicated that their technology can still identify people.

Facewatch from the UK announced that there was one Algorithm that can handle detection and identification based on only a person’s eyebrow region. The company proposes its retail store technology, saying that development beyond masks will extend to other covers, such as the religious veil called niqab, worn by some Muslim women.

Facewatch has already worked to identify people wearing hats and glasses, company spokesman Stuart Greenfield said. Customers, mainly retail stores, who wanted to keep shoplifters on a watchlist, did not consider mask detection until the pandemic to be particularly worrying.

“We only need the government we have to insist on [face masks]and the whole sector will have to respond very quickly, “said Greenfield. He added that Facewatch’s new algorithm will be able to identify people because their eyes and eyebrows are solid spots on the face and will change over time do not change.

Nevertheless, Facewatch expects some complications due to face masks. His algorithm usually identifies a person in half a second, and Greenfield said the masks could take longer. But the company said it was doing everything to make the new algorithm effective.

“Everyone is working to make sure we are fit for the market,” said Greenfield. “Our future depends on having a product that works exactly.”

SAFR, which promotes its technology for use in schools, also says that its tools can handle face masks.

“Our algorithms are now being trained on images of people wearing face masks,” said Hess. Until recently, the masks were not very present in society, “so they weren’t really added as training dynamics before,” he said.

To train its algorithm, SAFR relies on a variety of photos of people wearing face masks, some photos it creates itself, and others that its employees have provided at the company’s request. According to Hess, the company trains its algorithm on different images to take into account differences in gender, race and age.

The accuracy rate of the tools is 93.5% when people wear masks, said Hess, but only under ideal conditions, for example when the motifs are shown on a high-quality photo with adequate lighting.

skepticism

It is unclear how accurate these statements about facial recognition bypassing masks actually are. And it will be a while before we get definitive answers.

On the 1st of May NIST announced that it would be testing To determine exactly how facial recognition is for people with face masks, add digital masks to the existing photo database. However, the tests are closed due to the pandemic and there is no indication of when they will resume.

Facewatch and SAFR have announced that they will send their respective algorithms to NIST if possible. Without the test, there is no way to effectively compare the accuracy with other face recognition companies.

Right now, people have to take the word of a company that its technology actually works despite face masks. Facial recognition specialists are skeptical.

Kate Rose is an expert in digital security and the designer behind it Contradictory fashion. She makes clothes to trick surveillance technology, such as clothes to fool license plate readers and masks to thwart face recognition.

Rose tests the effectiveness of the masks at home using open source facial recognition tools and examines how surveillance technology detects people.

Face detection is used to scan and capture many data points on a person’s face, e.g. B. How far apart the eyes are and what the structure of the nose and chin is. For identification, the technology compares the scanned face with an image that is already in the database – one that is unlikely to contain a face mask.

Rose has no doubt that facial recognition providers can only identify people by their eyes and eyebrows, but she said that in a real-world scenario, this might be ineffective.

“If you have perfect pictures of my eyes, you can safely bring them together,” said Rose. “But the real world offers this crazy variety of background and lighting – and these things make everything really difficult.”

With the entire face, there is a greater number of differentiators that the AI ​​can work with. If the features are reduced to eyes and eyebrows only, there are much more similarities that can lead to false alarms.

The face masks would also play a role, said Liz O’Sullivan, co-founder of the AI ​​surveillance company Arthur. The trained algorithm may identify a person wearing a blue mask, but may be tripped by the same person wearing a red mask.

“When it comes to image processing, so much depends on how it is used,” said O’Sullivan. “Most likely, you would need a data set that contains the same person with and without a mask from different angles and lighting conditions. It might be possible to achieve the same goal only with the masked and unmasked pairs, but the accuracy would not.” be so high. ”

This is a problem that SAFR encountered in its tests, Hess said, describing how face masks can differ from country to country. For example, most masks used in Europe are blue, while in Japan some skin-colored masks are included in the company’s data set. These deviations could confuse the system.

“There will be some masks that go undetected,” said Hess.

Face recognition has long had problems with accuracy, especially in people with skin color and women. Adding face masks further complicates the task.

COVID-19 has hit minority groups particularly hard, with “a disproportionate burden of illness and death” affecting their communities. according to CDC. More than 80 percent of the subpoenas issued by the New York City Police Department Violations of social distancing From March 16 to May 5, colored people were exhibited, according to the department.

Experts warn that facial recognition and mask errors are another problem that minority groups may have due to the pandemic.

“The similarity between many different types of people will increase,” said Rose. “We all like to think that we are very unique and distinctive, but you will likely find many people in a dataset with very similar eyebrows and eyes.”

Beyond COVID

This new ability could have lasting effects long after the pandemic ended. Because of the public health crisis, companies are pushing for identification that can deal with masks. However, this ability could later be used by the police to identify protesters with face coverings.

In January, House Oversight Committee members warned that once facial recognition was perfected, it could have a deterrent effect on freedom of expression and civil liberties. The police have already used Face detection to monitor protestsand if people can be identified despite wearing masks, it creates a new level of privacy concerns for people who exercise their First Amendment rights.

Both SAFR and Facewatch said that they would not have started using face masks so quickly without the pandemic. But with the spread of the masks, their customers’ demand has increased.

“It is possible that you would have progress that would not have been made without it,” said Rose. “We should be aware that this can be a flood that will raise all boats in terms of surveillance.”

The information contained in this article is for educational and informational purposes only and is not intended as health or medical advice. Always consult a doctor or other qualified healthcare provider if you have any questions about an illness or health goals.