Abstract
One critical role of color vision is to segment scenes into distinct surfaces. The mechanisms that allow patches of color to be unified into a percept of a single colored surface remain largely unknown. We used fMRI to identify regions in cortex that contain neurons responsive to colored surfaces. Subjects viewed a novel dynamic stimulus comprised of an array of 59 × 44 small adjoining square elements whose colors were updated every 100 msec. Upon each update, squares were randomly assigned to be either surface or noise. Surface elements were all a fixed gray level and noise elements were randomly sampled from a uniform distribution of gray levels between the monitor's minimum and maximum. The surface color and ratio of surface to noise elements varied between trials. In pilot behavioral work, subjects reported perceiving a uniform colored surface behind the dynamic noise at surface-to-noise ratios above approximately 40%. Subjects in the fMRI experiment viewed stimuli at four surface-to-noise ratios, chosen to be symmetric around the mean threshold transition point from the pilot behavioral data. Stimuli were presented once every 5 secs for a duration of 2 secs, and interleaved with presentations of a static version of the same stimulus, but at 100% noise. Stimulus conditions were ordered using an m-sequence. Neural responses for each surface-to-noise ratio were estimated using ordinary least-squares, and amplitudes of the estimated responses were computed by fitting functions to the estimated response time course. Primary visual cortex responded more strongly to noise than to surfaces; response amplitudes decreased as the surface-to-noise ratio increased. Conversely, later visual areas responded more strongly to surfaces than to noise; amplitudes increased as the surface-to-noise ratio increased. These later areas likely contain neurons that integrate color information across relatively large areas in the visual field.