The role of low spatial frequencies for the categorization of faces versus nonfaces
Ben Vermaercke, Hans Op de Beeck · Lirias (KU Leuven) · 2008
Humans and other primates are able to categorize complex images with great accuracy and speed. Given that the crude low spatial frequency information in images is processed faster than high spatial frequencies, this categorization, and especially the fastest responses, might rely heavily on low spatial frequencies. As a consequence, fast and efficient object categorization might be feasible when only low spatial frequency information is available. We investigated this possibility with a study in which we asked human subjects to respond according to whether an image contained a human face (Go-trials) or not (noGo-trials). The images were shown at a size of 24 visual degrees, with the average size of the faces in the image being 12 visual degrees in horizontal extent. This is approximately the size of a real face when viewed from a distance of 70 cm. Images were either unfiltered or filtered with a cut-off frequency of 1.0 cycles/degree. Subjects were able to categorize the filtered images with high accuracy (94% correct), which is only slightly lower than accuracy with the unfiltered images (99%). We found that most errors with filtered images are made for stimuli that were categorized more slowly in the unfiltered format. This suggests that slower responses with unfiltered images are partially based on high-frequency information. Analyses of the reaction times showed that responses were slower overall for the filtered images (332ms) than for the unfiltered images (315ms). This difference was present across the whole reaction time distribution, so more very fast responses occurred for unfiltered images than for filtered images. This suggests that even the fastest responses with unfiltered images are partially based on spatial frequencies above the range preserved in the filtered images. Overall, these results suggest that low spatial frequency information is a very reliable source of information for object categorization, even though higher spatial frequencies have some role even for the fastest responses. In addition, our findings provide a benchmark to compare with the face categorization performance in human subjects with low vision (which are legally blind), as well as in “non-visual” animals (e.g., rats).