Abstract
Successful processing of configural information is considered essential to face recognition. The face inversion effect, in which recognition of upright faces is better than inverted faces, supports this assertion. However, it is less clear whether configural information is crucial to our ability to discriminate between faces. Faces vary both featurally and configurally, but hair, eye, lip and skin color may change dramatically while configuration remains relatively constant. Configuration is useful for recognition, but discrimination between faces relies on isolating differences. Therefore, it may seem more appropriate to rely on face features rather than face configuration when discriminating between faces. If so, there might be no benefit of expertise when discriminating around normally configured faces and no inversion effect. To test this, we measured discrimination thresholds with upright, inverted and contrast-negated faces. Each condition included five test points on a continuum of compressed to expanded faces, centered on an average face. The test points included an extremely expanded, slightly expanded, normal, slightly compressed and extremely compressed face. Observers scrolled through the continuum of faces around each point until they selected a just noticeably expanded or compressed face. Results indicate no benefit for discriminability around the average face. Rather, discrimination was best around the expanded faces, and became progressively worse with increasing compression for the upright and inverted but not contrast-negated faces. With the contrast-negated faces, performance was best around the extreme distortions and thresholds peaked around the normal configuration. The fact that there was no dip in discrimination thresholds around the normal configuration suggests that humans are not experts at discriminating between normally configured faces, and therefore probably rely on featural differences. Furthermore, the results showed no inversion effect: observers' discrimination thresholds for upright faces did not differ significantly from the thresholds for inverted and contrast-negated faces.
This research was supported by NIH grant EY01711.