Face Recognition in Challenging Situations

A great deal of previous research has demonstrated that face recognition is unreliable for unfamiliar faces and reliable for familiar faces. However, such findings typically came from tasks that used ‘cooperative’ images, where there was no deliberate attempt to alter apparent identity. In applied settings, images are often far more challenging in nature. For example multiple images of the same identity may appear to be different identities, due to either incidental changes in appearance (such as age or style related change, or differences in images capture) or deliberate changes (evading own identity through disguise). At the same time, images of different identities may look like the same person, due to either incidental changes (natural similarities in appearance), or deliberate changes (attempts to impersonate someone else, such as in the case of identity fraud). Thus, past studies may have underestimated the applied problem. In this thesis I examine face recognition performance for these challenging image scenarios and test whether the familiarity advantage extends to these situations. I found that face recognition was indeed even poorer for challenging images than previously found using cooperative images. Familiar viewers were still better than unfamiliar viewers, yet familiarity did not bring performance to ceiling level for challenging images as it had done in the cooperative tasks in the past. I investigated several ways of improving performance, including image manipulations, exploiting perceptual constancy, crowd analysis of identity judgments, and viewing by super-recognisers. This thesis provides interesting insights into theory regarding what it is that familiar viewers are learning when they are becoming familiar with a face. It also has important practical implications; both for improving performance in challenging situations and for understanding deliberate disguise.