A Multiple-Systems Approach in the Symbolic Modelling of Human Vision

McIndoe, Thomas (1992) A Multiple-Systems Approach in the Symbolic Modelling of Human Vision. PhD thesis, University of Glasgow.

Full text available as:
[img]
Preview
PDF
Download (32MB) | Preview

Abstract

For most of the thirty years or so of machine vision research, activity has been concentrated mainly in the domain of metric-based approaches: there has been negligible attention to the psychological factors in human vision. With the recent resurgence of interest in neural systems, that is now changing. This thesis discusses relevant aspects of basic visual neuroanatomy, and psychological phenomena, in an attempt to relate the concepts to a model of human vision and the prospective goals of future machine vision systems. It is suggested that, while biological vision is complex, the underlying mechanisms of human vision are more tractable than is often believed. We also argue here that the controversial subject of direct vision plays a crucial role in natural vision, and we attempt to relate this to the model. The recognition of massive parallelism in natural vision has led to proposals for emulating aspects of neural networks in technology. The systems model developed in this work demonstrates software-simulated cellular automata (CAs) in the role of mainly low-level image processing. It is shown that CAs are able to efficiently provide both conventional and neurally-inspired vision functions. The thesis also discusses the use of Prolog as the means of realising higher level image understanding. The symbolic processing developed is basic, but is nevertheless sufficient for the purposes of the present. demonstrations. Extensions to the concepts can be easily achieved. The modular systems approach adopted blends together several ideas and processes, and results in a more robust model of human vision that is able to translate a noisy real image into an accessible symbolic form for expert-domain interpretation.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Additional Information: Adviser: John Barker
Keywords: Computer science, Artificial intelligence
Date of Award: 1992
Depositing User: Enlighten Team
Unique ID: glathesis:1992-74810
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 27 Sep 2019 16:04
Last Modified: 27 Sep 2019 16:04
URI: http://theses.gla.ac.uk/id/eprint/74810

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year