Comparing higher-order spatial statistics and perceptual judgements in the stylometric analysis of art

Understanding the factors that underlie human perception of artistic style necessarily depends on measuring features beyond simple, low-order statistics. Indeed, much of our perception of style is rooted in qualities of lines and shading that cannot be described using first-order measures. In order to provide a richer, more accurate description of human perception of style, we must employ higher-order statistical methods. We demonstrate the applicability of two types of higher-order representations of images - and features derived from these - to the problem of similarity-based image search in large collections of art images. Our preliminary results indicate that a combination of perceptual information and statistical representations of art could prove extremely useful in navigating large art image databases in the context of similarity-based search.