Studies of the human visual system provide many important principles and models on which to base the design of machine vision systems. This thesis examines the application of such principles and models to the task of two-dimensional pattern recognition. Models for various stages of the visual pathway are considered: the log-polar model for the peripheral sampling structure of the retina, the mapping of this structure to the visual cortex, edge detection, the brightness constancy principle, centre of gravity fixation, and the Gabor elementary signals model for the response of simple cells in the visual cortex. Using these models, an algorithm is developed that is able to recognise well separated, two-dimensional, grey-level patterns of arbitrary position, size, orientation, and illumination. By seeking to emulate natural vision systems we may avoid expending effort on heuristic approaches for which there is no guarantee of success.

A method is demonstrated for relating the pattern features measured by the computer to the observations of humans. This method, based on a knowledge elicitation technique pioneered by Kelly, could improve man-machine interaction in visual domains, and is proposed as a new means of investigating the link between low and high level visual perception in humans.

The development of software for an interactive image processing system is also outlined. The human interface for this software is a structured command language shell providing a range of image processing functions. This language is designed to encourage experimentation and facilitate the development of image processing algorithms. A unique feature of this language is the ability to form links between variables and commands so that commands can be automatically invoked when an associated variable is modified. This software was used to carry out the pattern recognition experiments described in this thesis.