Fig. 4.

Distinctness and size-tolerance of the face-feature representations in V1, OFA and FFA. a) The linear discriminant analysis t-values and the corresponding corrected p-values are shown for all pair-wise comparisons of the face-feature response-patterns in V1 (top panel), OFA (middle panel) and FFA (bottom panel). The training data was the data from the first measurement session and the testing was done on data from the second session. In OFA, all face-feature stimuli could be discriminated from each other (no blue squares in the p-value matrix). b) To test for size-tolerance of the feature representations, the linear discriminant analysis was performed by training the classifiers on the small face-feature response-patterns and testing the classifiers on the large face-feature response-patterns. The results are shown as in (a). OFA shows successful generalization of the face-feature discrimination across stimulus size (no blue squares in the p-value matrix), suggesting size-tolerant feature representations.