Computer Vision summaries

theorem, which we can intelligently interpret. Above this minimum, however, there is a wide range
of sizes of image detail over which we appear to be able to interpret readily, immediately and
consistently. For instance, a head is instantly seen as a head, whether it is so small that it is a
minimally resolved blob or it occupies several degrees of the visual field. This capability appears to
exist not only independently of size, but also independently, as well as in parallel with, interpretation
of any internal resolved details which may be present. One might speculate that such a facility is
necessary for a versatile image processor, bearing in mind the uncertainties of presented image size
in many visual environments. It is therefore worth addressing the question "How is this size
independent performance achieved?".

11.2. Some relevant observations on human visual function.
11.3. Some relevant practical observations concerning computer image processing.
11.4. Concepts for efficient multiple scale analysis.
11 .5. Practical implementation
11.6. Conclusions.

Chapter 12. Local contour analysis.

12.1. INTRODUCTION.

Thus far in this book we have looked deeply at methods of extracting local single pixel information
about fragmentary profiles by learning from human vision. It is now time to broaden our horizons
and begin to consider how to synthesise local structural information from these fragmentary data and
how to interpret it. Trying to learn from human physiology in this case is not so easy, for two
reasons. Firstly, the processes involved are much deeper down the visual information processing