Abstract
There is a striking cost in face matching performance when members of a sequentially presented pair of faces differ in contrast polarity (i.e., reversal of brightness), but no such costs are apparent when matching objects such as chairs (Subramaniam & Biederman, 1997). This result holds true even when the faces and chairs, on mismatch trials, are of equal similarity, as assessed by a Gabor-jet similarity measure. Unlike face matching, object matching can generally be accomplished by using differences in parts and nonaccidental shape properties that would be unaffected by changes in contrast polarity. Would object recognition remain invariant when such differences were not available? Subjects matched sequentially-presented images of a pair of smooth, blobby, asymmetric volumes generated from rotations of the harmonics of a sphere. Stimulus pairs on mismatch trials spanned a large range of similarity. The highly similar pairs were far more similar than the faces or Greebles in typical face or Greeble recognition experiments. The Gabor-jet similarity measure was highly correlated with both RTs and error rates on mismatch trials. Changes in contrast polarity had no effect on the matching of these stimuli, either for naïve subjects or for “experts” that had practiced for over 8,000 trials with images of positive contrast. The invariance to object contrast held throughout the range of similarity, for both expert and naïve subjects. Practice markedly lowered RTs and error rates so that the performance of the experts was not only lower after practice than earlier in practice but also significantly better than the controls on the very first trials with blobs of negative contrast. These results suggest that faces are special with respect to sensitivity to direction of contrast.