A computational model of two-dimensional line drawing interpretations of partially occluded patterns based on simplicity principle

The key question of perception is how we manage to get an accurate, unambiguous and phenomenologically complete perception of the real world from proximal stimuli which are generally ambiguous and sometimes incomplete. Given a pattern as a visual input, we usually interpret it in one form although, in general, many interpretations of the pattern are possible. A study of the perceptual preference of partially occluded objects in two dimensional line drawings is presented. Two types of interpretations are considered, mosaic and completion. The interpretation is based on global as well as local simplicity. Global simplicity is measured by an information-load based on Leeuwenberg's model of coding theory and a minimum principle. The problems arising from this model are discussed and a solution based on an accessibility criterion is elaborated. However, this criterion alone does not solve the problem of local effect phenomena. Interpretations based on local cue information are then examined, and the issue of global versus local minima is considered from a computational perspective. In conclusion, a machine model of preference based on both local and global considerations is proposed, and its results compared to the results of psychological experiments on perceptual preference.