MACHINE VISUAL-PERCEPTION BASED ON SYMBOLIC 2-DIMENSIONAL SHAPE-DESCRIPTION .1.

In this paper we introduce a way to represent two-dimensional shapes by combinations of adjacent rectangular structures. A shape, by definition, is an area that is surrounded by boundaries. A context free grammar is used to build a simple, approximate, hierarchical symbolic representation of each shape, based on the set of rectangular structures that represent that shape. The symbolic shape representation enables us to apply high level reasoning about the shapes using a knowledge based system. Other aspects of vision-color, motion, etc.-are not assumed to be factors. The research goal is to investigate a method that reduces the amount of information that represents the shape and to formalize it so that common methods of reasoning about shapes can be used in typical applications that deal with arrangement of shapes of general types. Examples of such applications include cloth cutting in clothes design and manufacture, interior design, packaging, etc. The subject is multidisciplinary in its nature. Visual arts, information theory, human vision research, and several components of artificial intelligence: computer vision, knowledge representation and expert systems, approximate reasoning, heuristics-all contribute to the development of this subject. The choice of a proper approximation method is important. It must neglect a considerable amount of information but still keep the fundamental and relevant information needed for effective reasoning, and be valid for a large variety of types of shapes.