Simulated Human Vision..... Ian Overington

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

Edge-based Region Segmentation

Besides detection of edges in images, a second important form of image analysis concerns the segmentation of the image into defined regions having acceptably similar properties. This is usually considered, in the simplest cases, to imply regions of adequately similar brightness and chromaticity. Secondary attributes of similarity may include regions of acceptably similar textural properties (examples of a few of the many possibilities are shown to the right). These secondary attributes are, however, very many and varied, so these will be excluded from discussion here (although they are discussed & demonstrated at some length in Chapter 16 of my book ‘Computer Vision ....’).

Remaining with the primary properties of brightness & chromaticity, in an ideal world - such as with posters etc. - changes in these attributes are normally sudden, while regions between the sudden changes are usually roughly constant in brightness & chromaticity (e.g., the image to the left). In such cases the problem of region segmentation is relatively straightforward. However, in natural (real world) imagery there are several factors which make the problem substantially more difficult. Most of these may be loosely grouped under the general title of 'shading'.