Abstract
Inversion of a face can be readily shown to interfere with the matching or recognition of that face. Often this disruption is attributed to the inability to employ a configural representation whereby the whole face can be assimilated and judged as an integrated representation. However, direct evidence for this explanation has been lacking. Configural effects are enabled by the overlap of large receptive fields (r.f.s) centered at varied positions throughout the face which serve to magnify the impact of small metric differences distinguishing similar faces (Xu et al., 2014). Subjects judged whether a briefly presented sequence of two similar computer-generated faces, both upright or both inverted, with an intervening mask depicted the same person. When different, the faces could differ in the one of three areas of the face–upper, middle, or lower–with changes such as the height of the eyebrows, the length of the nose, or the width of the mouth. Fixation for a given trial was centered on the eyes, nose, or mouth, as the position of the faces was vertically shifted from trial to trial (but not within trials) (Fig. 1). The different feature could thus be 0 (< 1°), 1 (1–2°), or 2 (3–4°) positions, from fixation. For upright faces, there was little effect of distance from fixation in detecting a differing feature; for inverted faces accuracy fell with distance of that feature (Fig. 2). Accuracy on different trials, for both upright and inverted faces, were highly correlated (rs ~ .95) with the magnitude of the difference in the faces as scaled by the Gabor-jet dissimilarity measure (Fig. 3), with the benefit of dissimilarity greater for the upright faces. The greater cost of distance in inverted faces can be interpreted as a diminished benefit of a configural representation enabled by large r.f.s.
Acknowledgement: Harold W. Dornsife Research Fund