6 Knowledge Representation for basic visual categories

This paper reports work on a model of machine learning which is based on the psychological theory of prototypical concepts. This theory is that concepts learnt naturally from interaction with the environment (basic categories) are not structured or defined in logical terms but are clustered in accordance with their similarity to a central prototype, representing the "most typical' ' member. In attempting to operationalize this theory in a machine learning programme it has been necessary to study object descriptions of visual data. This has led to the development of a knowledge representation structure based on the natural properties inherent in visual images, shape and parts. Methods developed to achieve this are discussed. The organisation of knowledge derived from this approach differs fundamentally from existing K:R formalisms in leading to a view of concepts as perceptual structures at the basic level,with relations between concepts being described via a superordinate level of abstract concepts