Abstract
Behavioral studies indicate that scene categorization draws heavily upon analysis of scenes' spatial properties, such as three-dimensional layout (Greene & Oliva, 2009) as well as the kinds of objects scenes contain (Biederman, 1972; Davenport & Potter, 2004; Joubert et al., 2007). fMRI studies have identified distinct regions of ventral-temporal (VT) cortex that appear to process these features, notably the parahippocampal place area (PPA) for scenes' spatial properties and the lateral occipital complex (LOC) for their object contents. Although activity patterns in these regions correspond to the identities of viewed scenes (Walther et al. 2009, MacEvoy & Epstein, 2011), the extent to which perceptual judgments of scene category are determined by neural activity in these areas has remained unclear. This is particularly true of LOC, disruption of which has been shown to improve scene categorization accuracy even while degrading object recognition (Mullin & Steeves, 2011). To directly measure the influence of VT areas on scene category judgments, we used fMRI to record patterns of brain activity while observers categorized computer-generated scenes as bathrooms or kitchens; each scene was either unambiguously a member of one of these categories by virtue of both spatial properties and object contents, or was configured to be completely category-ambiguous. Observers were periodically instructed to base their decisions on either scenes' spatial properties or object contents. A classifier trained on responses to unambiguous scenes successfully predicted observers' categorical judgments of ambiguous scenes from the multivoxel activity patterns evoked in LOC and PPA. Crucially, predictions by LOC patterns were more accurate when observers were instructed to base judgments on scenes' objects, while those by PPA were more accurate for judgments based on scenes' spatial properties. These differentials are inconsistent with LOC and PPA patterns simply following observers' judgments, and instead indicate a significant dependence of judgments on neural signals in these regions.
Meeting abstract presented at VSS 2014