Abstract
The perception of surface reflectance has typically been restricted to the case of smooth matte surfaces viewed in simple illumination. In the real world, many surfaces are neither smooth nor matte, and they are viewed in complex illumination. Complexity can sometimes make the problems simpler; for instance, it has recently been shown that the statistics of natural illumination give rise to image statistics that can be used in estimating the reflectance of smooth objects like spheres. We have taken up the case of rough surfaces such as stucco, and have asked whether simple low-level image statistics might be used by humans in estimating their reflectance. We took photographs of rough materials and distorted their luminance histograms. Subjects were asked to rate the reflectance (black to white, on a five point scale) of a surface in the image whose mean, variance, and skew were independently manipulated by means of histogram matching to variable Beta distributions. Skew had a strong effect, as did the mean, while the variance did not. With the same mean and variance, the surface whose luminance histogram was more positive was systematically judged to be less reflective. The same experiments were run with phase scrambled versions of the images, and in the cases where these images were seen as surfaces, the histogram statistics had similar effects. These results support the notion that simple image statistics can be important factors in the perception of surface reflectance.