A graph-theoretic approach to colour image segmentation and contour classification

The segmentation of colour pictures in a graph-theoretic context is considered. The procedure aims at identifying visually important features on the image plane, such as regions of homogeneous colour and chromatic transitions. A segmentation algorithm of the region growing type is described which operates in a given colour co-ordinate system under the influence of an appropriate colour distance measure. The suppression of artificial contouring, a specific type of artefact introduced by the segmentation of algorithm, is formulated as a dual graph-theoretic problem. A hierarchical classification of contours is obtained which facilitates the progressive elimination of the undesirable contours. >