The design and application of associative memories for scene analysis

Abstract

This thesis investigates a novel scene analysis system that determines the identity and the relative positions of unconstrained objects within a natural three dimensional grey scale image. Images may be of 'block filled' or 'line drawn' occluded shapes. It utilises the occluding information to discover the relative depth of objects in the scene. The system incorporates associative memories, the N tuple pattern recognition process, movable multiple resolution windows and edge detection. The structure and performance of the system and its subsystems is reported. The associative memory incorporates a novel recall procedure which has uses outside the application given here. The work incorporates ideas from the neurophysiology of the human visual system to overcome some of the problems encountered.