The perceived beauty of art is not strongly calibrated to the statistical regularities of real-world scenes

Aesthetic judgements are partly predicted by image statistics, although the extent to which they are calibrated to the statistics of real-world scenes and the 'visual diet' of daily life is unclear. Here, we investigated the extent to which the beauty ratings of Western oil paintings from the JenAesthetics dataset can be accounted for by real-world scene statistics. We computed spatial and chromatic image statistics for the paintings and a set of real-world scenes captured by a head-mounted camera as participants went about daily lives. Partial least squares regression (PLSR) indicated that 6-15% of the variance in beauty ratings of the art can be accounted for by the art's image statistics. The luminance contrast of paintings made an important contribution to the PLSR models: paintings were perceived as more beautiful the greater the variation in luminance. PLSR models which expressed the art's image statistics relative to real-world scene statistics explained a similar amount of variance to models using the art's image statistics. The importance of an image statistic to perceived beauty was not related to how closely art reproduces the value from the real world. The findings suggest that beauty judgements of art are not strongly calibrated to the scene statistics of the real world.