Face spaces obtained from a reference PCA model, perceptual similarity judgments and fMRI response patterns in human visual cortex.

<p>(<b>a-b</b>) We generated 12 faces in a polar grid arrangement on a 2D slice through the reference PCA space (3 eccentricity levels and 4 directions). The grid was centered on the stimulus space origin. Euclidean distances between the 12 faces are illustrated in a distance matrix (a) and in a 2D visualization of these distances (b, multidimensional scaling, metric-stress criterion). (<b>c-d</b>) Group-average perceptual face space (N = 10) estimated from psychophysical similarity judgments. The distance matrix depicts the percentage of trials on which each pair of faces was rated as relatively more dissimilar. (<b>e-h</b>) Group-average cortical face-space (N = 10) estimated from fMRI response patterns in human visual cortex. The distance matrices depict a cross-validated estimate of multivariate discriminability where 0 corresponds to chance-level performance (leave-one-run-out cross-validation, <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1005604#sec017" target="_blank">Materials and methods</a>). Each cortical region was defined in individual participants using independent data. The distance matrices use separate color scales since we expect variable effect sizes across cortical regions. For data from other regions of interest, see <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1005604#pcbi.1005604.s002" target="_blank">S2 Fig</a>.