What Makes a Polygon Pleasing?

Abstract

Extensive preference data has been collected on Birkhoff's set of 90 polygons; therefore, they make an excellent test of any theory of aesthetics. None of the prior methods of explaining these data, however, are motivated by a theory of visual processing. Moreover, they do not provide an account of the response differences between artistically-trained and control subjects. The proposed method, based on a neurally-inspired model, attempts to correct these deficiencies. It is first argued that by measuring classification activity, one can gauge the extent to which the classical desideratum of unity in diversity is met. A classification model is then described in which this measure can be applied. It is based on the idea that recognition of an object can be thought of as a process of building up the object from its constituent parts. In this case, lines are built from localized line detectors and line-end detectors, polygon parts are built from lines, and the polygon itself is constructed from the parts. It is shown that the activity measure applied to this network correlates strongly with the preference data for the polygons for which the artists and controls agree. It is then demonstrated that the preference data for the polygons on which the artists and the controls differ can be accounted for by the effect of exposure on the network. Finally, it is argued that modeling preference data can provide a method of complementing other investigations into the nature of visual processing.