Importance of color in the segmentation of variegated surfaces

We examined how variations in color and brightness are used by the visual system in distinguishing textured surfaces that differed in their first- or second-order statistics. Observers viewed a 32 x 32 array containing two types of square elements differing in chromaticity or luminance or both. The spatial distributions of the two kinds of elements were varied within the array until observers could distinguish two juxtaposed regions. At low but not at high contrast, observers are better able to distinguish regions when the elements differ only in chromaticity than when they differ only in luminance. The advantage of color at low contrasts results from the greater visibility of the arrays defined by color variation. An observer's capacity to distinguish textures defined by variations in first-order chromatic statistics is little affected by the addition of achromatic noise but is more affected by the addition of chromatic noise. The relative robustness of chromatic cues in the face of achromatic noise leaves the visual system well equipped to exploit color variations in segmenting complex scenes, even in the presence of variations in brightness. This capacity seems to depend on mechanisms that sum over large regions: When surfaces differ in their second-order statistics and cannot be distinguished by mechanisms that sum over large regions, the advantage of color is much diminished.