|
Chapter 1. Introduction.
The topic of computer vision has produced many and varied texts over the last few years. One
must ask at the outset, therefore, whether there is really any room for yet another sizeable book. In
attempting to answer this question I have to admit that I am a relatively recent member of the
computer vision community, having spent many years attempting to unravel the main mechanisms
controlling human visual threshold performance. As a result of this latter background I have
frequently felt unable to accept that the computer vision techniques widely discussed in the popular
literature are remotely capable of achieving the levels of performance so readily achieved by human
beings. It has been argued "Why should one expect computer vision to achieve similar performance
to human beings?". On the other hand it has equally been argued "Why should computer vision be
limited to the performance limits found to apply to human vision?". I have formed very firm
convictions, from my wide studies of human vision and the visual environment, that there are very
good reasons why general computer vision (as opposed to systems designed especially for certain
very specific situations) should be limited roughly to the limits found in human vision. Perhaps the
most compelling of such reasons is that a majority of tasks for which we attempt to employ computer
vision are similar to visual tasks which have previously been carried out by human beings (and
normally carried out very well!). At the same time I equally firmly believe that computer vision
should be able to achieve performance close to those human limits with presently available optical
and computer hardware. My main purpose in writing this book is, therefore, to present, under one
cover, a wide variety of inter-related ideas and facts which have been collected together, both during
my work on human visual performance limits and my more recent work in the field of computer
vision. These collected ideas and facts together make what I consider to be a simple yet efficient
and unified schema for the front end of a composite computer vision system. I make no apology for
the fact that much of what I shall discuss is early processing, since I feel certain that, if one can
|
