Art statistics and visual processing: Insights for picture coding

Artwork holds much important information regarding the efficient encoding of the natural world, and it is therefore useful both for researchers in vision science and those in signal processing. Painters, like photographers, aim to capture the visual environment in a way that is appealing to viewers. But until recently, little attention has been paid to statistical regularities related to artists' representational strategies. How do artists deal with the very large dynamic range of luminances in scenes, when paintings themselves have a far smaller dynamic range? To what extent do artists reproduce the scale invariant spatial statistics of natural scenes, and what statistical regularities of natural scenes, if any, are retained when artists paint abstractly? This paper discusses findings that shed light on these questions and it suggests ways that these findings could spawn novel strategies for picture coding and image retrieval. It also describes links between artists' representational strategies and neural coding in visual systems.