Recognizing Faces Despite Variability in Appearance: Learning Mechanisms are Largely Intact in Older Adults

Abstract

Recognition of unfamiliar faces is highly error-prone, especially across changes in appearance (e.g., hairstyle, expression, lighting). Despite a lifetime of experience perceiving faces, older adults demonstrate poorer performance than young adults on unfamiliar matching and face recognition tasks. However, past studies have used tightly controlled images and so examined image recognition, rather than face recognition per se. No study to date has examined older adults' ability to recognize unfamiliar faces despite natural variation in appearance or the process by which older adults become familiar with newly encountered identities. We tested older adults (n=57) on a battery of tasks. First, we verified that older adults were highly accurate at recognizing multiple images of a familiar face (95% on a familiar card sorting task). To investigate the efficiency with which older adults learn new faces, participants performed a recognition task after learning three new identities—one from a single image, one from a low variability video captured on a single day, and one from a high variability video filmed over three days. Unlike young adults (Baker et al., 2017), older adults only showed evidence of learning in the high-variability condition (p=.002). This is consistent with evidence that children need exposure to more variability than adults to new a face (Baker et al.). To investigate whether older adults' inefficient learning is attributable to deficits in underlying mechanisms, we examined their ability to use ensemble coding (to rapidly extract an average representation of an identity) and to benefit from viewing multiple images in a perceptual identity-matching task. The results from both of these tasks suggest that these mechanisms are intact; like young adults, older adults show evidence on ensemble coding (ps< .001) and benefitted from viewing multiple images of a new identity (p< .001). Taken together these results have implications for models of perceptual expertise. Meeting abstract presented at VSS 2018