Abstract
In 2002, Graham Hole showed that, somewhat remarkably, vertically stretching a face by a factor of two (Fig. 1) had only a minimal effect on its recognition, as assessed by judgments of whether the face was that of a celebrity. This is surprising given that such stretching produces a marked change in the magnitude of the relations between the parts as well as on the shape of the parts and the head. Subjects viewed grayscale images of celebrities and non-celebrities either not stretched or stretched vertically by a factor of 2 (2x) or 4 (4x) and judged whether the face was famous or not. We confirmed Hole's results showing no effect of stretch on either RTs or error rates, but extended the evidence for invariance out to 4x stretch. Subjects varied in their familiarity with particular celebrities allowing a test of the possibility that the invariance to stretch was a consequence of having viewed familiar celebrities at various orientations in depth, perhaps producing familiarity with affine transformations of that face. This hypothesis was not supported in that there was no interaction between degree of stretch and the subject's familiarity with a given face. An alternative hypothesis is that humans have a general capacity for affine transformations of faces, independent of familiarity.
Meeting abstract presented at VSS 2018