Abstract
Responses to binocular disparity are widespread throughout the visual, parietal and ventral cortices, but what features are represented in these different areas? Here we use human fMRI and psychophysics to probe the nature of depth representations in the dorsal and ventral visual hierarchies. We presented participants (N=8) with slanted surfaces depicted in random dots stereograms (±7.5 to 52.5 deg in 15 deg steps) where stimulus size in the image plane was held constant as slant was varied. Psychophysically, we investigated two perceptual properties of these stimuli that varied with base slant: (i) slant discrimination thresholds and (ii) thresholds for the apparent physical size of the surface. We found that sensitivity to slant differences was highest at low base slants. In contrast, sensitivity to changes in surface size was highest at large base slants. Our fMRI measurements aimed to identify visual areas that mediate these perceptual results. During scanning, participants performed a demanding fixation task unrelated to slant or size judgments. We analyzed fMRI data using multi-voxel pattern classification analysis, and characterized decoding performance for neighboring slants in independently-localized regions of interest. We find a dissociable pattern of decoding performance in dorsal and ventral cortical areas. Specifically, sensitivity to neighboring slants was most pronounced for low baseline slants in area V3A and declined as base slant was increased. In contrast, decoding performance in area LO was highest for large slants, and declined as slant was reduced. These different patterns of response matched psychophysical judgments of (i) precise slant and (ii) surface size respectively. Our results suggest perceptually-relevant disparity representations in both the dorsal and ventral visual streams; however, the functional role of these representations appears to differ. Behavioral performance is compatible with accessing dorsal representations to make precise slant judgments, while ventral representations to compute object properties.
Meeting abstract presented at VSS 2012