Geodesic distance and MST based image segmentation

In this work, a new approach is proposed for the integration of spatial proximity information in graph based segmentation algorithms. This is done by means of the geodesic distance. Distance calculation and the implementation of the method are carried out using the minimal spanning tree (MST), constructed on a watershed image partition. Distance, defined over the MST edges, presents a measure of both spatial and feature coherence. It is incorporated in MST based color image segmentation applications, by means of a new density feature, which is computed with spatial locality restrictions.