Poster F31, Tuesday, March 28, 8:00 – 10:00 am, Pacific Concourse
Neural representations of face identity across photos, line drawings, and caricatures
Constantin Rezlescu1,2, Stefano Anzellotti3, Alfonso Caramazza1; 1Harvard University, 2University College London, 3MIT
Previous neuroimaging studies have identified brain areas involved in discrimination of face images (Kriegeskorte et al, 2007) or face identities across images varying in expression (Nestor et al, 2007) or viewpoint (Anzellotti et al, 2013). Stimuli used in these studies were real or computer-generated photographic images, so it is possible successful classification of identity was helped by specific combinations of lower-level features (e.g. a particular skin texture and color). We investigated if the standard face-selective brain areas (fusiform face area - FFA, occipital face area, superior temporal sulcus – STS, anterior temporal lobe) encode more abstract representations of face identity, not directly mappable to low-level features. To this end, we selected standardised photographs of three famous actors, varying in viewpoint (frontal, side, profile), and asked an artist to create sketched line drawings and caricatures based on those. The photographs, line drawings, and caricatures varied dramatically in their low-level features. Thirteen participants were scanned while performing an identity task based on these images. We trained linear support vector machine classifiers to discriminate between neural patterns in our regions of interest (ROIs) to specific identities based on two image types (e.g. photos and line drawings) and then tested how well these classifiers performed on the third, un-trained image type (e.g. caricatures). We found above chance classification accuracy in STS (9mm ROI) and FFA (6mm ROI). Our results suggest these areas contain more abstract information about face identity than previously shown. It is possible this information goes beyond visual properties.
Topic Area: EMOTION & SOCIAL: Person perception