Computational Selection of Processing Locations in Vision

Roger Browse

Introduction

Some of the issues addressed in computational vision research relate to the question of the structure of human intelligence in general, and are thereby subject to a Cognitive Science approach. This is particularly true of the traditional issue of how knowledge of specific objects might be applied towards visual recognition and understanding.

There are two advantages to the development of computer vision systems which maintain compatibility with selected aspects of human vision. First, clues to the underlying operational requirements of the vision system may be obtained from the characteristics of the human system, and second, the resulting computational system may act as an explanatory model for the human operations. The research reported here centers around a computational vision system which interprets line drawings of human-like body forms. This system maintains a number of compatibilities with human vision (see Browse 1981; 1982b).

This paper describes a mechanism for the interpretation-based integration of information available at different levels of resolution. From the integration, there follows a method for the intelligent selection of processing location within the image.