Minimal sorts for interpreting pictures

Computational approaches to simulate human beings interpreting pictures are important for understanding perceptual Gestalt and for building computer systems that support visual communication. Based on the minimum principle, we present a new approach to the interpretation of pictures. Our contribution is that we developed a novel way in which the geometrical information is calculated. In our approach, geometrical shapes are divided into various sorts and the sorts are organised into a hierarchical structure. A sort together with a number of points determines an actual graphical object. So, the objects themselves can be represented by the combination of their sort with certain points that code their position in the field and whatever other attributes they may possess as members of the sort. The geometrical information load is calculated as the number of points which are needed in the representation. Pictures are represented as a set of graphical objects. There are no other requirements on the input pictures. As long as the objects in a list are well-formed terms, interpretation can start. An inference mechanism reduces the terms in the list into terms which have the lowest information load. The deduced list of objects is the interpretation of the picture.