Abstract

Over the last twenty years a number of studies have analysed the image statistics of artworks to test whether the mark-making choices of artists can be described, at least in part, in terms of certain mathematical rules governing image content, and human aesthetic responses to that content. Debate in the field is still ongoing, and is driven partly by three unresolved questions:
(i) How do different statistical measures compare?
(ii) Are artistic choices influenced in some way by the values of particular visual statistics?
(iii) Do the values of particular visual statistics bear any relation to viewer responses to artworks?
This research attempted to address the questions by analysing 476 Western artworks dating from 1435 to 2008, drawn from the JenAesthetics and MART datasets. Results showed that:
(i) There are moderate correlations at best between the values of three different luminance statistics (Fourier spectrum slope, fractal dimension, and Shannon entropy), so they cannot be considered as equivalent and measuring the same image properties;
(ii) Statistical values are relatively stable over time and over art genres until the advent of abstract art in the early 1900’s.
(iii) Across the full image set there is no clear and simple relation between image statistics and viewer responses.