Painting What We See: Painters Over-Regularize Structure in the Environment

Abstract

We have recently investigated the relationship between the statistical properties of art images and biases in visual perception with a variety of manipulations (Schweinhart, Kim, & Essock, JOV 2011; Schweinhart, Williams, Dubinchik, O’Keefe, & Essock JOV 2012; Schweinhart & Essock, Perception 2012). These investigations are based on the supposition that when an artist paints a scene, we might expect it to contain the same structural regularities found in natural images in order to look "right" to the viewer. It is well documented that artworks contain the same regularities of scale (a spatial frequency fall-off of 1/frequency) as natural scenes (e.g., Simoncelli & Olhausen, Annu. Rev. Neurosci. 2001; Graham & Field, Perception, 2008; Graham, Friedenberg, & Rockmore, VisCog 2010). In our work, we investigate the presence in art images of another regularity of typical natural scenes-- on average, scenes contain a "horizontal effect" pattern of content: most content at horizontal and least at the obliques (Hansen & Essock, JOV 2004). Surprisingly, we find that artists actually over-emphasize the typical natural regularities in their paintings, particularly of orientation.. In average natural scenes, the anisotropy diminishes with increasing spatial frequency, but in paintings, the typical anisotropy is present at all spatial scales. Here we present findings from three projects: comparing museum paintings of landscape and portraits to photographs of faces and natural scenes; comparing paintings to photos of the actual scene painted; and a set of commissioned paintings completed under controlled conditions. All of these painting types indicate over-regularization in the representation of orientation. We suggest that this over-regularization is due to anisotropic suppression in the visual system that whitens the physical anisotropy and that artists capitalize on this in their artworks. In addition, we consider methods to remedy the problem of unequal sampling across orientations at low spatial frequencies in a FFT. Meeting abstract presented at VSS 2013