Abstract
Many of the phenomena associated with face (vs. object) recognition can be understood in terms of a representation for individuating faces that retains aspects of the original spatial filtering, as posited by Malsburg's Gabor Jet model (Biederman & Kalocsai, 1997). Objects, in contrast, may be represented by a structural description specifying explicit relations among view-invariant properties of edges of simple parts. To test whether the representation of faces, but not objects, retain characteristics of the original spatial filtering, subjects matched faces and blobs in a two-alternative, match-to-sample task. The blobs were smooth, asymmetric volumes (harmonics of a sphere resembling teeth) that, like faces, varied in the metrics of their surfaces. Each stimulus was filtered by a jet of Gabor wavelets at 5 scales and 8 orientations, with each jet positioned at the vertices of a 10*10 grid. On half the trials, the correct choice was the identical image as the sample; on the other half it was a complement of the sample. Complementary pairs of images were produced by assigning every other scale and orientation component to one member of a pair and the remaining components to the other. Consistent with the hypothesis that face representations specify the original spatial content, matching complements of faces resulted in greater error rates than matching identical images for both novices and experts. No such costs were apparent when matching blobs, a result consistent with prior findings in the Fourier domain with faces, chairs, and blobs. A pixel-based ideal observer analysis showed that faces and blobs had equivalent complementary costs, indicating that the greater cost in matching complementary images of faces compared to blobs was not due to intrinsic differences in the stimuli.
This work was supported by Human Frontier Science Program RG0035/2000B, NSF 0426415, NSF 04207994. Thanks to Shuang Wu, Ken Hayworth