Face masks and fake masks: the effect of real and superimposed masks on face matching with super-recognisers, typical observers, and algorithms

Since the outbreak of COVID-19, mask wearing has been required in various settings. It is important from a theoretical and applied perspective to understand the impact of face masks on people’s ability to recognise faces, and this has consequences for security/identity verification purposes. To date most research investigating the impact of face masks on face recognition has not used real images of people wearing masks, but has superimposed a mask image on to a pre-existing face image. This is true for research using humans as well as computers, and in fact the world standard test of algorithms uses superimposed face masks (https://pages.nist.gov/frvt/html/frvt_facemask.html) Here we ask whether the literature could be underestimating the problem posed by face masks through its use of superimposed (fake masks) as opposed to images of wearing face masks. In face matching tasks using real and superimposed masks, we show that super-recognisers and control participants, as well as algorithms, perform less accurately with real masks than superimposed masks. Therefore, by superimposing masks on to pre-existing stimuli we may be underestimating the problem they pose for face identification.