Purchase this article with an account.
David R. Andresen, Kalanit Grill-Spector; Mixture of view-invariant and view-dependent representations in human object-selective cortex. Journal of Vision 2005;5(8):743. doi: 10.1167/5.8.743.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Humans recognize familiar objects from almost any viewpoint (i.e., view-invariance). However, behavioral invariance need not imply that underlying neural representations are also invariant. Indeed, the nature of object representation is under heavy debate. We used fMRI-adaptation to investigate sensitivity to changes in viewpoint using a parametric manipulation of in-depth rotation from an initial adapting view. Critically, this method allowed us to determine the degree of invariance as a function of rotation and adapting view.
Eight participants completed ten runs of adaptation and test in a 3T scanner. During adaptation, participants viewed 10 line-drawings of animals and vehicles repeated six times while performing a 1-back matching task. For five runs participants were adapted with front views of objects, and for the other five they were adapted with rear views. At test, they viewed objects rotated in-depth 0 (identical), 60, 120, and 180 degrees, as well as new objects. Participants categorized each image as animal or vehicle.
Surprisingly, responses in object-selective cortex were largely invariant (i.e. remained adapted) to changes in viewpoint when adapted with front views of animals, but were view-dependent (i.e., showed a systematic recovery from adaptation with increased rotation) when adapted with rear views of the same animals. In contrast, responses in these regions were view-dependent when adapted with vehicles, regardless of the adapting view.
These results suggest that representations in object-selective cortex are largely view-dependent, but the information available in a particular view affects the level of invariance. Responses to informative views (e.g., front views of animals) are more invariant than less-informative views (e.g., rear views of animals). These findings suggest a mixture of view-invariant and view-dependent representations exists within object-selective cortex, and provide critical constraints for theories of object recognition.
This PDF is available to Subscribers Only