Abstract
It has been suggested that the perception of gloss in natural material images is strongly related to the skewness of the luminance histogram (Motoyoshi et al., Nature, 2007). Here, we set out to evaluate the validity of this findings in a large and diverse material image dataset of 1492 images. The images were chosen from several material image databases, encompassing 10 different material categories. 8 subjects were asked to categorize each briefly (33 ms) presented image as either glossy or matte by the press of a button, or, if undecided, not give an answer at all. Subjects were encouraged to respond as quickly as possible. Only responses given within 500 ms were taken account. Images were sorted according to the degree of consistency with which they were assigned to one of the two groups. 76 images were consistently rated as matte, and 112 images were consistently rated as glossy by all 8 observers. Crucially, we found the two sets of images to be significantly different, with glossy images being substantially more positively skewed. Images could correctly be classified as "glossy" or "matte" with an accuracy of 75% based on the skewness parameter. This difference became systematically smaller when we considered images with ratings of decreasing consistency. Our results agree with earlier findings by Motoyoshi et al (Nature, 2007) and Anderson & Kim (Journal of Vision, 2009). Even in a very large and varied stimulus set, glossiness and skewness are associated. However, skewness alone is not sufficient to induce the percept of gloss in each single image.
Meeting abstract presented at VSS 2013