Surgical face masks may impair facial identification

People may not be able to correctly verify the identity of individuals in photos if they are wearing surgical face masks, according to a study published in the open access journal Cognitive Research: Principles and Implications.

London, 19th November 2020

People may not be able to correctly verify the identity of individuals in photos if they are wearing surgical face masks, according to a study published in the open access journal Cognitive Research: Principles and Implications. Facial recognition systems may still be able to function accurately, but should be updated, according to the authors.

Since the beginning of the COVID-19 global pandemic, surgical face masks, typically worn by healthcare professionals, have become a popular choice of face covering in response to government guidelines recommending or requiring citizens to wear face coverings in public.

To investigate the accuracy with which people can verify the identity of individuals wearing a surgical face mask, researchers at the University of Stirling asked 138 participants to view pairs of photographs of either familiar celebrities or unfamiliar models. Participants were randomly assigned to one of three groups and shown two faces on screen simultaneously; both wearing a mask, only one wearing a mask, or neither wearing a mask. Images of the surgical masks were superimposed onto each photo using editing software. The participants were then asked to decide whether the pair of images showed the same person or two different people.

The authors found that participants were less likely to correctly match faces wearing surgical masks, regardless of whether one, or both faces, were masked. Surgical masks caused the same decrease in performance for both the familiar celebrity faces, and the unfamiliar models – a finding that surprised the authors. If shown two unfamiliar faces, participants were more likely to report a “mismatch” when either one or both faces were masked, than if both faces were shown unmasked. Participants were more likely to report a “match” if they recognised at least one of the faces in the photographs and if either one or both faces were masked, compared to if both faces were unmasked. The authors suggest that this bias towards reporting a match for familiar faces may partly be explained by participants comparing the features of the masked face to images of the celebrity previously stored in their memory.

Dr Daniel Carragher, the lead author of the study, said: “Our study demonstrates that surgical face masks significantly impair our ability to identify both familiar and unfamiliar individuals, and that the impairment is similar whether one or both faces in each pair are masked. This study differs from previous work, because the participants could always see the two faces they were comparing – just like a cashier that asks to see your photo-ID at the store. We think this may be because face masks might disrupt the ability of the observer to process the face as a whole. But, it is also possible that adding a mask to a face may also alter the perceived appearance of the top half of the face, which typically has a larger influence on face recognition than the lower features.”

Overall, the findings suggest that future research could focus on designing transparent face coverings to reduce the spread of disease, while still allowing identification of the individual underneath.

The study also investigated whether surgical face masks would impair the performance of facial recognition systems that had not been trained to recognise masked faces. They found that the main system used in this study was able to extract enough information from the top half of a masked face to outperform most human participants at accurate identification. However, the study also shows that face masks cause significant impairment to the performance of several other commercially available face recognition systems. The authors caution that not all facial recognition systems can be used to accurately identify masked faces, and that extensive testing of those that attempt to do so is necessary before they are used in the field.

Further Information

About the journal: Cognitive Research: Principles and Implications.

Services for Journalists

The full-text article is available here.

Contact

Kirsty Bone | Springer Nature | Corporate Affairs
tel +44 (0)20 3192 2286 | kirsty.bone@springernature.com