How do masks affect facial recognition solutions?

 

Facial recognition technology has seen major growth and a lot of hype in the security industry, and beyond, in recent years. However, as wearing a face mask has become common practice in countries worldwide since the outbreak of Covid-19, there is an obvious question: 

 

How do masks affect the performance of facial recognition solutions?

 


The problem

 

Facial recognition solutions identify a person by forming a unique code built on algorithms from multiple points on a person’s face, including nose, chin, lips, eyes and jaw. However, when a person wears a mask, many of these key points are not visible.

 

Major issues

 

IPVM, a leading video surveillance testing and research authority, recently carried out research on the performance of facial recognition solutions on masked faces. They tested four solutions which produced conclusive results:

 

"In our tests, facial recognition confidence dropped dramatically (over 50 points confidence) when wearing anti-viral face masks, with subjects not recognized at all in most cases, and no faces even detected much of the time,” states the article.

 

Faces were often completely missed, and unsuccessful or false identifications were high. Further tests revealed that a subject with a combination of a mask and another obstruction, such as hair, sunglasses or a hat, made recognition near impossible. The technologies simply had too few points on the face to work with. 

 

They concluded that since most widely used solutions are not currently trained to recognize masked faces, the scenario posed major challenges for facial recognition technologies across the board.

 

Successful cases

 

However, several Chinese companies now boast the ability to successfully carry out facial recognition on masked faces.

 

Chinese and Hong Kong brands, such as Hanwang, SenseTime and Gato, have developed solutions which produce accurate results with masked faces by retraining their technology to only work with data from around the eyes and forehead.

 

However, these solutions are generally designed to work in office settings with a limited number of people in the database (up to 50,000 employee faces in the case of Hanwang’s solution). Moreover, other optimal settings in an office building scenario, such as good lighting and the subject presenting their face to the camera, makes it easier to produce good results.

 

For facial recognition deployed in public areas or and where there is a database of millions of people it is much more difficult to achieve the same accurate results.

 

The bottom line

 

Most current facial recognition solutions will encounter problems with masked faces. 

 

Although there are solutions on the market which have overcome this challenge, and more may be developed with further training of algorithms, mask poses a fundamental problem for facial recognition: the technology has fewer points to work with. Accurate recognition is therefore more difficult and risk of false identification is higher. 

 

However, technology is constantly evolving and improving, so this story doesn't end here. Moreover, other biometric solutions are gaining traction in the market, such as gait recognition which identifies people by their body shape and walking posture. If wearing masks continues to be norm, we may see an increase in the development and deployment of this technology and others.

 

 


 

 

If you are considering adopting facial recognition in your security solution and would like further advice on this topic, reach out to us at solutions@icdsecurity.com.

 

 

Find out more