Abstract
Sandford and Burton (2014) asked participants to rescale faces to normal after being initially presented with faces of distorted proportions. The important and surprising result was that participants were better at normalising unfamiliar faces compared to familiar faces. They suggested this was due to an increased tolerance to distortion with familiar faces: a result interpreted as questioning the role of relational information in familiar face recognition. We repeated their study in two experiments. In both experiments participants first rated our set of faces for familiarity on a 7-point scale. For each participant, only faces rated as very familiar (7) or not familiar at all (1) were included in analysis, though all participants viewed all faces in rating and normalising tasks. In addition to manipulating familiarity (familiar or unfamiliar), orientation (upright versus inverted) was also manipulated. Experiments 1 and 2 showed an effect of orientation on normalising error with upright faces normalised more accurately than inverted faces. Neither Experiment 1 nor 2 showed the previous influence of familiarity in normalising error. Experiment 1 actually showed familiar faces to be normalised more accurately than unfamiliar faces, in contrast to the result of Sandford and Burton. In Experiment 2, participants also completed the Cambridge Face Matching Task (CFMT: Duchaine & Nakayama, 2006) and Cambridge Face Perception Task (Duchaine, Germine & Nakayama, 2007). Correlating test performance with error in normalising faces showed a relationship between performance in the Cambridge Face Matching Task and normalising faces. Better face memory was associated with more accurate normalisation for familiar and unfamiliar, upright and inverted faces. We consider the effect of orientation and the failure of our two experiments to replicate the Sandford and Burton finding, alongside the correlation with performance on the CFMT. We suggest a likely mechanism underpinning the ability to normalise familiar and unfamiliar faces.
Meeting abstract presented at VSS 2017