Purchase this article with an account.
Stefano Anzellotti, Scott Fairhall, Alfonso Caramazza; Decoding orientation-invariant information about individual faces in the ventral stream. Journal of Vision 2012;12(9):1179. doi: https://doi.org/10.1167/12.9.1179.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In order to recognize individual faces we need to distinguish very similar images (specificity) while also generalizing identity information across image transformations such as changes in luminance and rotations in depth (invariance). The neural mechanisms that underlie the ability of preserving identity information across image transformations remain unclear. We investigated the presence of orientation-invariant information about faces applying information-based mapping to the BOLD signal in the ventral stream, training a pattern classifier to discriminate different faces seen from 4 orientations in depth, and testing its performance at discriminating those faces seen from a fifth novel orientation. Rather than showing a gradual increase from the posterior to the anterior ventral stream in the presence of informative voxels for orientation-invariant classification, we find that informative voxels cluster at the level of the fusiform and at the level of the anterior temporal lobes. To investigate the presence of orientation-invariant information in standard regions of interest, we defined ROIs for the OFA, FFA and right ATL using an independent localizer. Highly localized orientation-invariant information about faces was detected in the right ATL, where patterns of activity extracted from a small number of voxels allowed to achieve cross-viewpoint face classification significantly above chance. Orientation-invariant classification of faces was also achieved in FFA and OFA when considering larger numbers of voxels, while classification in V1 remained at chance. These results suggest that representations in the FFA and OFA encode visual details about the faces, while representations in the right ATL seem to abstract away from such details and can therefore be contained in a smaller extent of cortex.
Meeting abstract presented at VSS 2012
This PDF is available to Subscribers Only