Abstract
It has previously been shown that image statistics, similar to those used for texture analysis, can be useful in recognizing optical qualities such as gloss. The results are strongest for smooth objects of known geometry. We now consider the case of complex surfaces of unknown geometry such as crumpled paper. We find that textural statistics are quite informative here as well. We analyzed images of natural materials (paper, cloth, powder, vegetables) and found that the statistics were highly dependent on the lightness and color of the material. This is apparently due to the influence of interreflections and transmission within the surface itself (related to the effects in Gilchrist's black and white rooms, and Langer's colored rooms). As a result, it should be possible to do a certain amount of lightness and color constancy by looking at a single isolated surface, without comparing it to a surrounding context. I will describe a number of perceptual phenomena that are consistent with this expectation, and which indicate that humans use textural statistics for much more than just texture analysis.