The Fusiform Face Area responds automatically to statistical regularities optimal for face categorization

Abstract

Statistical regularities pervade our perceptual world. Assuming that the human brain is tuned for satisfying the constraints of the visual environment, visual system computations should be optimized for processing such regularities. A socially relevant and highly recurrent homogenous pattern for which the brain has developed sensitivity is certainly the human face. Yet, for which statistical regularities the face sensitive regions are tuned for, and to what extent their detection occurs automatically is largely unexplored. Using fMRI we measured activations within the face sensitive areas for nonface symmetrical and asymmetrical curvilinear patterns with either more high-contrast elements in the upper or in the lower part. Faceness evaluation performed outside of the scanner showed that these patterns were not perceived as schematic faces. Noticeably, symmetry violations disrupted perception of faceness, despite objective image similarity measures showing high faceness values for those patterns. Among the faces sensitive regions, only the right Fusiform Face Area (FFA) showed sensitivity to symmetry. This region showed also greater responses to patterns with more elements in the upper part. Critically, the FFA's responses were more strongly correlated with the physical objective faceness properties of the stimuli than the perceived subjective faceness ratings of the observers. These findings provide direct evidence that the neural computations of the right FFA are tuned to curvilinear symmetrical patterns with high-contrasted elements in the upper part, which fit best with the physical structure of human faces. Such low-level geometrical regularities might be used by the FFA to automatically categorize visual shapes as faces.

Figure 1

Left: examples of stimuli used during and after the fMRI scanning. Right: the face‐like control stimulus.

Figure 2

Image similarity means and their standard errors between the stimuli used for the fMRI experimental conditions and the face‐like control stimulus (cosine ([0,1]). The average image of the patterns in each respective condition is reported inside the bars. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 3

Normalized faceness rating scores, and their standard errors, for the fMRI experimental conditions (0: minimum; 1: maximum). Note, the ceiling effect for the face‐like control stimulus that was not presented during the fMRI scanning sessions, indicating that the schematic face‐like pattern was perceived and rated as the most face‐like stimulus by all the participants. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 4

Functional regions of interest for faces identified by the localizer paradigm in one participant showing the three ROIs (R/r = right; L/l = left). The images are displayed in radiological convention so that the left direction on the page corresponds to the right direction on the anatomy and vice versa. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 5

Mean BOLD signal percent change measures across participants, and their standard errors, related to the processing of the experimental visual patterns within the functionally defined brain ROIs. The relative number of participants presenting significant activations for faces within their respective selective regions is reported in parentheses. [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

Figure 6

Top‐right: Average face calculated across the images of 35 men and 35 women presenting 7 expressions (neutral, fear, anger, disgust, happy, sad, and surprise—KDEF face database, 1998). The composite image results in a symmetrical face, with a triangular top‐heavy bias. Bottom‐right: Composite image resulting from averaging face composites presenting different views. Note, that now symmetry disappears, as this property is not reliably shared across the exemplars. Critically, however, the final face composite still presents more elements in the upper part (red dots). [Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]