Abstract
Specular highlights on typical object surfaces reflect spectral components of illumination directly, and thus may be a cue for estimating illumination color. Some previous studies reported small improvement in color constancy in scenes with glossy objects. Here, we psychophysically investigated if this highlight effect on color constancy can be explained by low-level image statistics of glossy objects. The stimulus was composed of a test sphere and many background objects rendered with Ward model. The specular reflectance of the background objects was the same across the objects in a scene and was either of five levels from 0 to 0.2. In contrast, diffuse spectral reflectance of each background object was fixed at either of those of four Munsell color chips. The illumination was either of D65, A, or 25000K. The test sphere’s specular reflectance was fixed at zero, and its chromaticity can be controlled by observers. In addition, we had the control noise condition, in which background objects were replaced by noise images created by phase randomization of the original objects and with the same luminance and chromaticity histogram as them. Observers made achromatic setting on the test sphere. In the results, roughly speaking, color constancy improved with specular reflectance monotonically in accord with previous studies. However, this improvement in color constancy was much smaller in the noise condition. These results suggest that improvement of color constancy by adding specular reflection components cannot by fully explained by differences in low-level image statistics between glossy and matte objects.