Basic Contents

Computer Vision ... (Published 1992)

Simulated Human Vision..... Ian Overington

Location: Eastbourne. UK
ianoverington@simulatedvision.co.uk ............ www.simulatedvision.co.uk

input to the threshold performance model ORACLE when the stimuli being viewed were complex. It
was soon realised that the transformed data at the optic nerve contained some ambiguities,
particularly for small sizes and fine detail. It became clear that other important transformations of
which we must take account were those which took place in the early stages of the visual cortex,
particularly in area 17 of the striate cortex (see also Chapter 2). Rather slowly, therefore, over the
following few years, studies were carried out to extend VISIVE to include these extra processes. A
boost was given to the studies when it was realised that understanding and simulation of early visual
function may have an important bearing on automatic recognition systems. Thus in the early 1980's
there was a major effort to complete and refine VISIVE, such that it could be a viable option as a
preprocessor for artificial intelligence systems. It has been argued by the author that it is most likely,
after all the time which human vision has had to evolve, that the human visual system now utilises a
near-optimal series of processing steps for many of the image interpretation tasks for which machine
vision systems will be used. At the same time it has been observed that many facets of image and
information processing, which appear, from classical computer vision literature, to be quite difficult
processing tasks, are carried out essentially instantly and with apparently effortless ease by human
beings. The studies carried out to refine VISIVE, and subsequent studies to be described later in the
book have, we believe, amply confirmed that many of these 'difficult' processing tasks can indeed be
performed simply and efficiently by learning from human vision. At the same time they are
considered to provide compelling evidence that our interpretation of the maze of neural details in
human vision is relatively close to the truth. It is now possible, as will be discussed in this and the
next two chapters, to obtain a rich harvest of information, of very high spatio-temporal fidelity, from a
basically simple, but carefully specified, preprocessor. Further more, the data produced are of a
suitable form and fidelity to permit a wide range of subsequent image analysis processes to be
carried out in a unified and interactive manner.