Interpretation Based Interaction Between Levels of Detail

Roger A. Browse

Introduction

Interaction between different levels of detail (resolution or spatial frequency) play an important role in computational vision. For the most part, the operations involving these different levels do not relate to the scene domain knowledge being used in interpretation (Hanson and Riseman, 1974; Tanimoto, 1976).

On the other hand, Kelly (1971) and Shirai (1973) have shown that each level of detail may be separately interpreted and that the result of one level may aid the interpretation of the other. Rosenthal and Bajcsy (1978; Bajcsy and Rosenthal, 1980) describe how search for specific objects in an image can be assisted by being able to predict the level of detail at which an object can be located by exploiting the containment relation between scene domain objects.

This paper describes the operations of a computational vision system which permits interpretation of images through cooperative interaction among information obtained from more than one level of detail. It has been previously shown that this approach provided compatibility with aspects of human vision (Browse, 1981). This paper outlines the system's ability to reduce interpretation possibilities by providing scene to image domain mappings at two levels of detail.