hide
Free keywords:
-
Abstract:
The light reflected from a surface depends on the reflectance of that surface and the spectral power distribution of the incident light, thus making it impossible to predict surface color directly from its wavelength composition. Despite this computational problem, the human visual system is remarkably accurate at inferring the color of surfaces across different illuminants. This ability is referred to as color constancy and it is essential for the organism to use color as a cue in object search, recognition, and identification. We devised images of two surfaces presented under three different illuminants using physically realistic rendering methods to disentangle the influences of wavelength composition, surface reflectance, and illumination. Measuring patterns of fMRI voxel activity elicited by these images we tested to what extent responses to surface color in various retinotopically mapped visual areas remained stable across illuminants. While surface color could be decoded in all ROIs when the illuminants did not differ between training and test sets, we found generalization across illuminants in V1 only. When viewing the scene in a cue conflict condition that abolished color constancy as measured psychophysically, generalization also broke down in V1. When fMRI activity was elicited by stimuli that were matched in reflected light but differed in illumination and therefore also perceived surface color, higher visual areas showed an increasing bias towards surface color representation and a decrease in illuminant color representation. Our results demonstrate the differential roles that V1 and V4 areas play in transforming chromatic input into color constant percepts.
