Are you struggling to identify your neighbour’s face behind the mask? Well, you’re not alone! Computers are having a hard time too, breaking facial recognition algorithms as per the new government study.
A study circulated by the US National Institute of Standards and Technology (NIST) suggested that the most widely used and the best facial recognition systems show error rates up to 50% while trying to recognize masked faces.
Facial recognition finds it really hard to identify faces, especially in black masks than the blue ones.
Since the pandemic has begun, there is a need to fully understand the dynamics of this technology as to how it handles the masked faces, stated by Mei Ngan, a NIST computer scientist and a researcher of the report. Thus, we have now started the process by focussing on how our current algorithm that was created prior to the pandemic is affected by faces wearing masks. Thus, this summer, we have planned to test the algorithms’ accuracy, developed intentionally, after taking into account masked faces.
Considering this issue, Apple made it feasible for iPhone users to access their phone without face ID earlier this year. Authorities also find it hard to identify people participating in violence at the protest of Black Lives Matter and similar gatherings.
NIST claims that they will further investigate better to understand the function of facial recognition for covered faces. This research has studied only those algorithms produced prior to the pandemic. Still, their coming step is to assess how commercial agencies are going to improvise their technology according to the society where mask-wearing has now become a norm!
Image credit: CNET
Companies that are associated with law enforcement agencies have attempted to tailor the Facial recognition function by focusing on the eyebrows and eyes only instead of the whole face.
The researchers involved in the study examined the technology through virtually drawing masks in a series of international traveller photographs. They then matched those photos with the unmasked people’s database seeking visas and various other immigration benefits.
The research body claims that it scanned around 6.2 million images of about 1 million people using 89 algorithms provided by tech firms and academic labs.
For the top facial recognition systems, the failure rate stands at 0.3% under ideal condition. However, several studies indicated significant disparities in facial recognition systems across race, age and gender. With the addition of masks, the error rate surges to 5% or even worse. With covered faces, the agency states that several competent algorithms have failed in between 20 – 50% times.
Even in the pre-pandemic era, governments were still inclined towards recognizing concealed or covered faces.
For example, Masks became an official mark for demonstrators in Hong Kong, because it provided protection against tear gas and helped them to cover the face as a result of the fear of retribution in case they get publicly identified. However, the government had also implemented restrictions on covering faces.
Then we met the Coronavirus pandemic, and everything changed! Now health officials globally are strongly emphasizing people to cover their mouth and noses with a mask.
NIST’s study suggests that the accuracy rate of facial recognition systems is highly dependent on what and how people wear their masks. The more the face is covered, the more difficult it is for algorithms to identify the person behind a mask.