Abstract
To achieve color constancy, the visual system must estimate the illuminant. An influential proposal for illuminant estimation is to assume that the brightest element in a scene is either a white surface or a specular highlight and therefore provides the illuminant color. We tested an alternative hypothesis: Observers use the geometry of the surface and the illumination to select highlight regions, even when they are not the brightest elements in the scene. In computer-rendered scenes we manipulated the reliability of the "brightest element" and the "highlight geometry" cues to the illuminant, and tested the effect on performance in an operational color constancy task. To eliminate other cues to the illuminant, scenes contained only a single spherical surface illuminated by multiple point sources of light, each with the same spectral content. The surface reflectance took a single spectral distribution but was modified by surface texture that attenuated the reflectance by a variable scale factor. The surface had one of three levels of specularity: zero (matte), low and mid. In the experiment, observers saw a one-second animation and their task was to indicate if the color change was due to an illuminant change or a material change. Discrimination performance was close to chance for matte surfaces, as predicted. However, as specularity increased, performance significantly improved. Importantly, it was shown that performance exceeded the prediction given by an ideal observer using the brightest element to perform the discrimination. Moreover, separate analyses for trials in which the specular region fell on a dark part of the texture showed an additional performance enhancement, even though the brightest element heuristic would predict a performance decrease. These results suggest that human observers do not simply rely on the brightest element in constancy tasks, but rather utilize the geometry of specular regions to separate surface and illuminant properties.
Meeting abstract presented at VSS 2017