As restrictions continue to ease globally despite the surging numbers of COVID-19, one thing has left an unexpected impact: face masks. Face masks have been recommended to required in the past weeks, and have drastically changed the course of facial recognition AI.
Cameras and video analytics worldwide that have been installed in airports, schools, government buildings and on city streets were designed to identify us by our facial structure. Now, half of the data that they are intended to read has been remove by the application of a face mask. Scarves, bandannas, masks and other face coverings have been used by people during the pandemic to attempt to control the spread of the coronavirus.
To attempt to continue the progress of these programs, researchers have collected databases filled with photos of random masked faces. The purpose of these images is to run them through various programs to help improve facial recognition algorithms.
While many groups have worked to help cameras to recognize masks specifically, recently groups have been working to ensure that facial recognition can still progress as a whole during this time. In April, a post was published to Github containing a COVID-19 Mask Image Dataset that contained more than 1,200 images collected from social media. A month prior, researchers from China compiled a similar database of more than 5,000 masked photos from online also.
Further, Apple has begun to roll out updates to its software in an effort to make it easier to unlock the iPhone using facial recognition while wearing a mask. According to Cnet, Apple requires a user’s eyes, nose and mouth to be visible for the phone to be unlocked. Obviously, a mask interferes with that criteria.
While there has been no substantial progress made yet to relying on just the upper half of the face to identify someone, tech companies are constantly working towards that goal, and will only strengthen facial recognition algorithms for the future.