Abstract
In both the human vision and the machine vision communities, there is much interest in the statistics of natural scenes, especially the statistics of the esponses of wavelet-like filters. These statistics can offer useful information about surface properties such as albedo and gloss. Machine vision systems can learn these statistics and use them, for example, for separating eflectance from illumination. In many cases, by manipulating image statistics artificially we can change a surface's appearance. Skewness, which measures the asymmetry of a distribution, is of particular interest because it is correlated with gloss and albedo for a range of surfaces. We find evidence that humans use skewness, or something like it, in estimating surface properties. It is straightforward to measure skewness with physiologically plausible mechanisms.