Regularities in the Anisotropic Content of Portrait and Landscape Paintings: A Corollary to the Horizontal Effect Anisotropy of Visual Processing

Abstract

Structural content in natural scenes is biased in spatial scale (1/frequency) as well as orientation (H>V>>Oblique), and neural encoding appears to "undo" (whiten) these biases (Simoncelli & Olhausen, Annu. Rev. Neurosci. 2001; Essock et al., JOV 2009; Hansen & Essock, JOV 2004). Thus, when an artist paints a landscape, we might expect that it contains the same biases to look "right"--if it didn’t, the most prevalent scales and orientations would be over-emphasized by the anisotropic suppression. Recently, the relationship between the structure of natural scenes and paintings has been explored (e.g., Graham & Field, Perception 2008; Graham, Friedenberg, & Rockmore, VisCog 2010). Specifically, we have reported that artists actually overemphasize the anisotropies found in natural scenes in their paintings (Schweinhart, Kim, & Essock, JOV 2010). Here, we compared the statistical properties found in portrait paintings to natural scenes and photographs of faces. There is some indication (Redies et al., Network 2007) that the amplitude spectra of portraits are closer to those of scenes than faces, but the orientation spectra of portraits has yet to be studied fully. Paintings were photographed every 3o (avoiding a discrete transform bias; Hansen & Essock, 2004). The images were then analyzed by fast Fourier transform (FFT) and power at different orientations was compared across spatial frequency. Scenes show a strong horizontal-effect anisotropy (H>V>>Ob) that doesn’t vary with spatial frequency while faces show no anisotropy. Interestingly, paintings of both landscapes and faces show the anisotropy both with increasing magnitude and spatial frequency, even though the pattern is not present in all types of subject matter. Meeting abstract presented at VSS 2012