Is COVID affecting Facial Recognition? What do experts say?

By Rahul Vaimal, Associate Editor
  • Follow author on
Facial Recognition with Mask
Representational Image

In recent years, facial recognition used by governments and private companies has become more reliable and common, owing to deep learning, which is an artificial intelligence technique, that makes computers even better at processing images. But what effect has COVID-19 and Facial Masks have had on the effectiveness of the technology.

Countries like China, UK, and Russia are using them widely. Statistics suggest that China has one security camera for every seven persons.

The facial recognition technology and hence the industry was on a roll. It was expected that, by 2024, the demand for facial recognition would reach $7 billion. While, Norton, the internet security firm, expected it to reach $7.7 billion by 2022.

But then, the coronavirus shook the world.

Blocked by masks

One by one, the countries enforced strict measures to ensure the usage of mask usage to reduce the spread of the virus and the facial recognition technology firms looked on helplessly leaving them uncertain about the future.

While wearing a medical mask about half of your face gets covered. Also, this is the area that the camera depends on to gather most of its information. Biometric data is majorly collected from the central portion of our face.

How does face recognition work?

Initially, the system ‘detects’ a face in the scene. Next, it takes a snapshot of the individual and transforms it into comparable data based on various factors like the gap between the eyes, the curve of the lips, and distance from forehead to chin.

While basic systems are capable of calculating only a few facial features, the sophisticated ones calculate dozens of them. But all this information is useless without a reference image for comparison. A reference image may come from a number of sources like a passport document, for instance.

How are tech companies going to deal with masked faces?

But those old reference photos are no good when people show up with masks. So, tech firms have been trying to find a way around the problem.

They are trying to catch up as fast as possible through three techniques.

The first and simplest way is by gathering masked selfies for training facial recognition algorithms. Of course, there should no scarcity for these selfies as all of us have uploaded at least one of them

“The face mask selfies aren’t just getting seen by your friends and family, they’re also getting collected by researchers looking to use them to improve facial recognition algorithms,” say experts. It is reported that the researchers have been able to lay their hands on “thousands of face-masked selfies up for grabs in public data sets including Instagram.”

Another way to do this is to mask existing photos. For example, Chinese company Hanwang ‘s face recognition program “works for masked faces by trying to guess what all the faces in its current database of photographs would look like if they were masked,” says Ars Technica. Hanwang ‘s chief technical officer, Huang Lei, said their system ‘s accuracy with masked faces has increased from roughly 50 % to 95% in the lab.

Thirdly, they are developing systems that have the capacity to recognize faces just by using the area around the eyes.

Existing problem

China has one of the most advanced systems when it comes to facial recognition. Chinese citizens use their face to pay at a store or to take money out of an ATM.

Experts from China say masked face recognition is not a new challenge as their companies have been facing similar problems with people with northern Chinese customers who wear winter clothes that cover their ears and face. Similar is the case of people with beards in Turkey and Pakistan.

They say that “occlusion detection”, recognizing when a face is obscured — has been a feature of professional face detection systems for some time. Somebody wearing sunglasses, a scarf, a low hat, or with their face not turned directly toward the camera needs to be detected too.

Stuart Greenfield, a spokesperson for UK-based Facewatch, says that even before the pandemic his company has been focusing on detecting people wearing hats and glasses.

Focussing on the eyes

Eyes are a major source of data when it comes to facial recognition as eyes and eyebrows are fixed points on the face and don’t change over time and can do 75% of the job unless the person is wearing sunglasses.

Though a perfect picture of an individual’s eyes can do the trick in facial recognition, it is very difficult to procure them with varying backgrounds and lighting.

The approach to identify masked faces by scanning just the eyes and cheekbones is called ‘periocular recognition’. But mostly the system works better only when the subject is directly facing the camera. When they turn away, the number of false matches increases tenfold.

How reliable will be the future face recognition technology?

The companies developing these systems say they are accurate, of course, but are face recognition systems really ready to deal with masks?

Facial recognition technologies have many drawbacks even without masks as they can be confused by factors like aging, plastic surgery, makeup, effects of substance abuse or smoking, pose of the subject, and image quality.

Also, each individual is unique. Hence, person A may have unique features around the eyes, while person B has a unique mouth or jawline. If both are wearing masks, the machine can quickly recognize A but not B.

Experts say that ‘the technology’ can even trip on the color of masks. The deviations in the color of the masks have the potential to confuse the systems.

Further, experts are of the opinion that face recognition with masks will be harder in large populations. When the number of people is less than 100,000 at a time, a major difference may not be noticed but with 1 million people, accuracy may be noticeably reduced.

None of the proposed advancements seems to be as reliable as facial recognition from a fully visible face and the possibility of increased mistakes in crowded areas is worrisome. But the tech companies remain optimistic and are looking to make the most of the opportunity.

Show full news