An object Model for Image Recognition

Object-oriented data models have been developed for several image applications, such as top-down scene description, object and component matching, and representation of structural and spatial relations. In this paper we examine the use of an object-oriented model for the complementary problem, that of automatically building the hierarchical scene representation from pixels to higher level entities by means of image processing and pattern recognition. In particular, we discuss the issues associated with instantiating a hierarchical structure bottom-up, while viewing it top-down from the user's perspective.