Abstract
Accurate face recognition is easy for viewers who are familiar with the faces concerned, but highly error prone for viewers who are unfamiliar with them (Bruce, 1986; Burton et al. 1999). An influential proposal is that people become good at recognizing a face by learning its configuration—specifically, distances between facial features. Here we test this proposal experimentally using a natural manipulation of these distances. Harper & Latto (2001) showed that changing camera-to-subject distance also changes distances between features. However, their pioneering study did not test implications of these image changes for identification performance. Following Harper & Latto (2001), we photographed volunteer models at both Near (32 cm) and Far (270 cm) viewing distances, resulting in changes to inter-feature distances in the image, confirmed by anthropometry (Kleinberg et al. 2007). Experimental participants who were either Familiar (N = 22) or Unfamiliar (N = 23) with these faces viewed pairs of images in a matching task that required Same Identity or Different Identity judgments. Images were paired to create Same Distance (i.e. Near+Near or Far+Far) and Different Distance (Near+Far or Far+Near) conditions. Familiar viewers performed accurately in both the Same Distance condition (M = 99%, SE = .29) and the Different Distance condition (M = 97%, SE = .89). In contrast, Unfamiliar viewers performed much more poorly in the Different Distance condition (M = 81%, SE = 1.42) than in the Same Distance condition (M = 91%, SE = 1.04). The finding that Familiar viewers were impervious to these non-linear changes in facial configuration suggests that familiar face recognition is not strongly dependent on distances between features in the face image.
Meeting abstract presented at VSS 2015