Abstract
Two face images of identical contrast superimposed in transparency with the same orientation in the image plane (e.g. both upright) are usually perceived as a single fused image of a new identity. However, when the two faces in the transparency display do not share the same orientation, one of the two dominates perception over time. It is usually the image whose orientation is closer to upright that has the highest dominance or salience, being perceived for longer in alternation with the weaker image, or as having a higher subjective contrast. If the two images are identical (i.e. same person) but have different orientations (e.g. upright vs. inverted) it is possible to equalize the salience or perceived contrast of the two components by reducing the physical contrast of one while increasing the other. We have thus obtained orientation tuning functions for face processing by measuring the contrast ratios at which observers judged the two components to be equally salient for each possible orientation pair. The tuning functions can be described as bell-shaped curves tapering off at orientations of 90 degrees from upright. To control for possible effects of lighting we have used sets of faces weakly lit from above and from below, obtaining in both cases very similar results. Finally, we have estimated with the same technique the tuning curve for images of scrambled faces, in which the internal features had been rearranged to appear in an unnatural position. In this case observers gave virtually identical judgements for all orientation pairs, showing no evidence of a preference for any of the orientations of the image. We conclude that human face processing has a narrow orientation tuning (i.e. ±90 degrees) centered on upright, which is not dependent on lighting direction, but is contingent upon the normal facial configuration being intact. Our results have a striking similarity to the tuning obtained for a population of cells in monkeys' STS.