Abstract
A number of recent studies have compared the statistical regularities of natural scenes to those of paintings (e.g., Graham & Field, Perception 2008; Graham, Friedenberg, & Rockmore, VisCog 2010). The theory behind such investigations is that art may reflect coding biases inherent in human vision since art is created to be “effective” when viewed by the visual system. Several studies (e.g., Graham & Field, Spatial Vis 2007; Redies et al., Network 2007) have shown that structural content in art work show a bias of scale that is similar to that of natural scenes – a linear fall-off of amplitude with spatial frequency. However, natural scenes also tend to be anisotropic with most power at horizontal, next most at vertical, and least at the 45° obliques (e.g., Hansen & Essock, JOV 2004). Furthermore, a corresponding anisotropy (the “horizontal effect”) appears to be inherent in the coding of orientation by the visual system (Essock et al., JOV 2009). A comparable anisotropy in paintings may depend on scene content and painting style, but representational paintings of natural scenes, and perhaps other types of content, would be expected to contain this bias. The current investigation seeks to compare orientation anisotropies found in natural scenes to the orientation content in certain classes of paintings. Paintings were photographed in several museums. To avoid bias associated with different orientations, the camera was rotated obtaining images every three degrees. A Fast Fourier transform (FFT) was performed in Matlab on the images and average amplitude extracted at 0° and 90°. From this set of image transforms, the full 180° of content was obtained (see Methods in Hansen & Essock, 2004). For most painting types, results confirm prior reports of “1/f” structural relationship and also support our conjecture of a horizontal effect anisotropy.