Abstract
The ability to rapidly decode the 3D structure of a scene is a crucial ability for navigation and survival. To accomplish this, the brain has to translate two-dimensional images on the retina into three-dimensional percepts. Human depth perception relies on both binocular and monocular visual cues. Here, we investigated neural processing of monocular depth cues, produced by a textured ground plane, using multivariate pattern analysis of fMRI data. In separate scanning sessions, subjects were presented with two stimulus configurations: lines arranged either such that they formed a texture gradient leading to a depth percept of a receding ground plane, or randomly distributed such that no ground plane was perceived. Both stimulus configurations were present in each session, but line orientation differed between the two sessions. Thus there were four stimulus types: ground planes defined by vertical or horizontal lines, and scrambled vertical or horizontal lines equated for low-level image properties. We performed cross-categorical classification by training on lines of one orientation, and testing on the other. This was done to identify areas distinguishing the presence or absence of a ground plane, without relying on low-level stimulus differences. A searchlight analysis revealed a network of areas where it was possible to classify the presence of a ground plane independent of the orientation of the stimulus elements, including clusters in the superior and middle temporal and frontal cortices, as well as fusiform gyrus. Unlike within-orientation classification, cross-categorical classification did not occur in early visual cortex, which suggests that it did not rely on image-level features. This conclusion was further supported by an ROI-analysis revealing that cross-classification was not possible within V1, V2 or V3. These results support the existence of a wide-ranging network of brain areas dedicated to processing depth information derived from monocular cues.
Meeting abstract presented at VSS 2012