Abstract
The top half of one face appears to fuse perceptually with the bottom half of another, when aligned spatially and presented upright. While this composite face illusion (CFI) is easy to demonstrate, there is considerable disagreement about how best to measure observers' susceptibility. Prevailing matching approaches require observers to discriminate target regions in pairs of composite faces. However, findings differ depending on the authors' choice of design ('standard' or 'complete') and their approach to analysis. Here we present a novel CFI paradigm based on psychophysical approaches to the measurement of illusion susceptibility. Observers made binary judgements about target regions ('male' or 'female'; 'child' or 'adult'). Target regions were drawn from morph continua, either blending a male and a female face, or a boy's and a man's face. In the gender variant, distractor regions were unambiguously male or female; in the age variant, the distractor regions were unambiguously child-like or adult-like. In this procedure, susceptibility to the CFI biases observers' classification performance; for example, observers are more likely to classify targets as male or female in the presence of aligned male and female distractors, respectively. Susceptibility to the CFI is inferred from the degree to which an individual's psychometric function shifts in the two conditions. When aligned with targets, distractor regions induced substantial shifts in both the gender and age variants. As expected, misaligned distractors and inverted arrangements yielded little or no shift. Function shifts for age and gender were correlated in 67 typical observers suggestive of a common binding process. However, shifts failed to correlate with face recognition ability measured using the Cambridge Face Memory Test. Moreover, 16 developmental prosopagnosics showed typical shifts when tested on the gender variant. Taken together, these findings challenge the view that susceptibility to the CFI is closely related to face recognition ability.
Meeting abstract presented at VSS 2017