A neurocomputational account of the magnitude of face composite effects.

Identical top halves of two faces are more likely to be perceived as different when their different bottom halves are aligned rather than offset. Here, we demonstrate that the magnitude of the offset effect for each face can be predicted from a model of overlapping receptive fields with tuning profiles similar to the hypercolumns of simple cells in V1, although the cells are likely in face-selective areas. Importantly, a single face part (e.g. the left eye) is coded by multiple large receptive fields centered at a distance from the face part (Fig. 1). When different bottom halves are aligned to the identical top halves of faces, the large receptive fields centered on the top half of the face will extend to the differing bottom halves, thus making the top halves of the faces more dissimilar. By offsetting the differing bottom halves from the identical top halves of two faces, the features of the bottom halves no longer activate the large receptive fields centered on the top half of the face, leading to more accurate judgments of the identical top halves as the same (Fig. 2). The retention of early-level visual coding (Yue, Tjan, & Biederman, 2006; Xu, Biederman, & Shah, 2014) and the retinotopic representation of a face template in FFA (de Haas et al., 2014) may explain why the offset effect is unique to faces rather than the parts-based representation of objects. Meeting abstract presented at VSS 2015