Predicting aesthetic experience with spatial and chromatic image statistics

<p dir="ltr">This thesis examines relationships between chromatic and spatial image statistics in aesthetic judgements and aims to understand whether aesthetics is calibrated to real-world scene statistics, and how relationships with image statistics vary between individuals. Paper one examines judgements of photographs of common objects in their real-world context. Partial least squares regression explained 26% and 36% variance in aesthetic preference and perceptual naturalness respectively, and these two variables were significantly correlated. Paper two presents three experiments probing individual differences in the contribution of image statistics to aesthetic judgements. One experiment found that levels of autistic traits, as measured with the Autism Quotient (AQ), moderated the relationship between some image statistics and aesthetic preference for real-world photographs. In subsequent experiments, we showed that the moderating effect of AQ was absent for judgements of computer-generated fractal and chromatic patterns. Cluster analyses revealed substantial individual differences in aesthetic response profiles to variation in fractal dimension and chromatic distribution, but clusters were unrelated to AQ. Overall, the thesis finds relationships between low-level chromatic and spatial image statistics and aesthetics that are not strongly calibrated to real-world scenes but vary across individuals. Paper three examines whether beauty judgements are calibrated to the scene statistics of real-world scenes. Results showed that image statistics did not predict beauty ratings of artworks better when expressed in units of deviation from typical values for real-world scene images taken by head-mounted cameras. However, beauty ratings did increase the more similar the chromatic distribution and edge density similarity of art is to real-world scenes. These findings have implications for the roles of efficient coding, natural scene statistics, and individual differences in aesthetics. The research demonstrates the predictive power of biologically plausible image statistics, paving the way for future investigations of the neural bases of relationships between image statistics and aesthetic experience.</p>