Adaptive image analysis for object recognition. I. Entropic object location

Abstract

We construct an automatic and adaptive image analysis method for machine vision after the Marr primal sketch model for human visual perception. In Marr's model, higher forms of visual perception and recognition are built upon a set of primitives which constitute a base set of features recognized by the eye-brain system upon viewing a scene. The human visual system is intrinsically adaptive. A challenge for any robust machine vision system is to retain this adaptive aspect of vision, and at the same time be able to recognize complex objects in an efficient manner. This paper, presents the framework for such a system based upon the fundamental information theory concept of entropy, the use of recursion, and a spatial object hierarchy permitting redundancy resolution. We describe our methods of identifying and locating an object within an image, including the use of multiple thresholds selected using a recursive algorithm and an image entropy function. Located objects are traced and methods for extracting interior object information are described.