Caricaturing Improves Face Recognition in Simulated Age-Related Macular Degeneration.

Age-related macular degeneration (AMD) damages central vision, leaving only blurred peripheral vision. This severely impacts face recognition and thus social functioning. Here, we explore whether face recognition could be improved in AMD, by manipulating the physical face image such that it better suits the perceptual mechanisms humans use to recognise facial identity. We enhance identity information in the face via caricaturing each individual face away from an average face, matched to the target face for sex, race, age, expression, and viewpoint to ensure enhancement specifically of identity information. For high resolution (unblurred) face images, caricatures are better individuated than veridical faces. We replicate this caricature advantage. We then simulate early- through late-stage AMD by filtering spatial frequencies to mimic the appearance of a face at approximately 10° through 30° into the periphery, for a face sized as if viewed on a tablet computer held in the crook of the arm. Results show that the caricature advantage is present for even the most highly blurred images. Specifically, pairs of different-identity faces presented simultaneously are rated as more dissimiliar with increasing degrees of caricature; and in a face learning task, old-new recognition is improved at least for new faces. Remarkably, for 10° blur, caricaturing improves performance to 'normal' levels (i.e., equal to veridical unblurred faces). Finally, we explore the origin of the caricature advantage within the visual system. Some caricature advantage is present for inverted faces, and in observers of different race to the target faces. This argues at least part of the caricature advantage arises from mid-level shape vision, or possibly feature-level coding of faces that is less sensitive to orientation and race than whole-face coding. We conclude that properties of multiple stages of the visual system may be relevant to choosing image manipulations that improve face recognition in AMD. Meeting abstract presented at VSS 2013