Simulated Human Vision..... Ian Overington

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

Edge-based Region Segmentation (cont.)

If one looks critically at virtually any real world scene, very little of it will consist of patches of uniform brightness & chromaticity. Consider, for instance, a version of the posterized image on the previous page which incorporates ‘natural’ shading as shown right - or the typical portrait image shown far right. In both these cases the ‘regions’ are of substantially other than uniform brightness & colour. Yet we accept - generally without any thought - that areas bounded by closed edges are individual 'regions'.

If conventional techniques for region segmentation (such as 'region growing') are applied in such circumstances, there is a distinct danger of problems if the shading is other than minimal. Hence we believe that human vision tackles the problem in a different way - by segmenting the regions as part of the edge finding process. In this way, both local edge characteristics (leading to local form perception) and local closed regions are integrally related.