Abstract
Parsing of visual scenes into objects and their surround is largely dependent on the ability to rapidly and accurately detect boundaries between adjacent regions of space. To assess the cortical mechanisms underlying figure and background processing and to evaluate the role of boundary cues in segmentation, we constructed displays in which figure and background regions were separately ‘tagged’ with periodic modulations of their local texture elements. Using these texture-defined objects and an electrophysiological paradigm that allowed us to monitor figure-region, background-region, and border-specific activity separately in fMRI defined visual areas, we have found distinct neuronal networks for the processing of each surface and the formation of region boundaries. Figure activity was distributed over a network of ventral visual areas including the lateral occipital cortex, while background activity extended from primary visual cortex through the dorsal visual pathway. Both figure and background activity was invariant with respect to the texture cues used to define the regions, their locations, and their spatial extent. Responses related to non-linear figure-ground interactions (measured at the sum of the two tag frequencies) involved both first-tier and extra-striate areas. This activity depended on the relative configuration of the regions, diminished monotonically with the inclusion of increasingly large gaps placed between the regions, and was absent in the presence of only a single region tag. Border activity is therefore necessary and sufficient to generate activity at non-linear harmonics of the regions tags, indicating that these frequency components provide an index of lateral contextual arrangements in spatial vision.
Supported by EY014536, EY06579 and the Pacific Vision Foundation