Facial recognition and face masks are a bad combination. Here’s a solution

Jack Turley August 26, 2020
Facial recognition and face masks are a bad combination. Here’s a solution

Facial recognition and face masks are locked in a war with no easy way out. The software is stumped by our current necessity for wearing face masks and, if you’ve got a recent iPhone, this can be a real nuisance. 

Many companies have jumped at an opportunity for innovation, claiming high levels of accuracy in facial recognition even with face masks on. 

However, a recent study by the US National Institute of Standards and Technology (NIST) has shown how facial coverings significantly increase error rates across the board. 

So, does this study prove that all the companies are talking waffle? Are some closer to a solution than others? And how can you avoid locking yourself out of your iPhone while wearing a face mask?

A scramble for facial recognition success

The coronavirus pandemic has made face masks a safety requirement in many everyday scenarios. First it was public transport, then shops, and now even schools are among the places where facial coverings are crucial for slowing the spread of the disease. 

Newer iPhone users realised this meant returning to the stone ages of actually typing in their phone password – or getting locked out of their device – when Face ID was stumped. 

Companies raced to find facial recognition technologies to counter this, trying to identify people based solely on their eyes and cheekbones

The latest NIST update has shown that, despite the claims, no company has yet achieved the same level of success as facial recognition without face masks. Nevertheless, some companies are definitely making progress.

A student reads while wearing a face mask. Photo credit: Pixabay

Increasing error rates

In July, NIST tested algorithms designed pre-covid and found masks caused erroneous results in 5-50 per cent of cases

With this latest report, there was some evidence of workable error rates. Dahua, a Chinese corporation, saw their algorithm failures rise from 0.3 per cent without a mask to only 6 per cent with one on. This still represents an overall working accuracy. 

However, many companies, like Rank One and TrueFace (used, among other things, to detect strangers in US schools) presented greater error gradients. RankOne saw their failure rate rise from 0.6 per cent to 34.5 per cent. TrueFace, which is also used at US Air Force bases, saw a mask-free vs mask-on error rate increase from 0.9 per cent to 34.8 per cent. 

Of the algorithms tested, all 41 showed lower rates of facial recognition success with face masks involved. Although the percentage difference was small for companies like Dahua, this problem is far from resolved. 

With health experts suggesting face masks will become the norm for years to come, further technological advancement is still required. 

The challenge of identifying a half-covered face is clear. Photo credit: Pixabay

An Apple solution for facial recognition with face masks

For users of the iPhone X and more recent models, this doesn’t mean you need to lament the absence of the home button until better facial recognition is available. Try this instead:

  1. Fold your mask in half. 
  2. Reset Face ID. 
  3. Create your first face scan holding the folded mask on one side of your face. 
  4. Repeat (using the Alternative Appearance option) with the mask covering the other side of your face. 

This method was developed by students at the Tencent Xuanwu Lab

China seems to be leading the way in all aspects of facial covering solutions. Perhaps this is due to masks already being in widespread use there prior to the coronavirus outbreak. 

So, have a go at the work-around above. Get in contact and let us know if it worked for you, or if you’ve found a better way to unlock your iPhone while wearing a face mask. 

Have something to tell us about this article?
Let us know
J E Turley is a novelist and freelance writer. To read more about his novels, or about living with Crohn’s and Colitis, head to jeturleywriting.co.uk.