Basic Contents

Computer Vision ... (Published 1992)

Simulated Human Vision..... Ian Overington

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

14.9. Practical colour image processing.
14.10. Conclusions.

Chapter 15. Edge-based Region Analysis.


For general scene analysis and interpretation, one important operation which must be carried out is
segmentation into bounded regions having similar properties. It is considered that this operation is a
necessary companion to virtually
any other scene analysis. This is equally true whether the scene
analysis problem is considered at a relatively subjective level, attempting to segment into regions
such as 'trees', 'sky', 'houses', 'vehicles' etc., or whether it is considered as the analysis of details of
small and select portions of a scene objectively. After all, I would argue that, although the statistical
properties of a portion of a scene can be
analysed by putting an artificial window around it, unless a
boundary can be perceived around a characteristic region it is very unlikely that one can be sure to
interpret the region correctly.

Much attention has been given to the general topic of scene segmentation in recent years, it being
usually assumed that segmentation should be in terms of local means and distributions of
levels. Whilst considerable success has been achieved by such techniques as region growing and
'split and merge', we have become increasingly aware that a large percentage of the structure and
form information in real scenes, at least as far as
human vision is concerned, is associated with local
and rather sudden
changes of energy level. If all the changes of energy level in scenes were
sudden, and if all sudden changes were related to segmentation boundaries, then the methods of
segmentation based on local energy levels might be expected to work very reliably. However, many
of the sudden changes in natural situations are related to texture or structure
within conceptual