Purchase this article with an account.
Vaidehi Natu, Fang Jiang, Abhijit Narvekar, Shaiyan Keshvari, Alice O'Toole; Representations of facial identity over changes in viewpoint. Journal of Vision 2008;8(6):159. doi: https://doi.org/10.1167/8.6.159.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Neuroimaging studies have established ventral temporal (VT) cortex as a brain region involved in human face processing. To be useful for face recognition, neural representations must operate robustly over changes in three-dimensional viewpoint. We carried out an fMRI experiment to examine the brain response activations for facial identities as they varied across viewpoints, and for viewpoint when facial identity varied. Five participants were scanned while performing a one-back task on faces presented in identity-constant blocks (faces varied in viewpoint) and in viewpoint-constant blocks (faces varied in identity). The stimuli consisted of four male faces viewed from four viewpoints ranging from the frontal (0 degrees) to the profile (90 degrees) in increments of 30 degrees. The four identities included two face-"anti-face” fairs (cf., Leopold et al., 1999), which were created to be maximally dissimilar from each other. We applied a pattern-based classifier to the task of discriminating brain response patterns for individual facial identities and viewpoints. Voxels used for the classifier were selected from VT brain regions found to be responsive to faces in a standard localizer scan session. The classifier was a linear discriminant analysis that operated on a representation of the scans projected onto their principal components. For each participant, classifiers were applied to the task of discriminating all possible pairs of identities and all possible pairs of viewpoints. Within the individual participants, moderate but above chance discrimination levels were found for facial identities and viewpoints, although discrimination patterns were not entirely uniform across the participants. These results suggest that it may be possible to apply pattern classification techniques to the complex task of discriminating the subtle neural codes involved in face representations.
This PDF is available to Subscribers Only