Segmentation of Dot Patterns Containing Homogeneous Clusters

M. Tuceryan and Narendra Ahuja

Abstract

The problem of segmenting a dot pattern into its homogeneous cluster components is considered. Geometrical features of the Voronoi neighborhoods of points are used as similarity measures to group points. A probabilistic relaxation labelling algorithm is designed to label each point as an interior or an edge point of a cluster, based upon the area and eccentricity values of the Voronoi polygons. Experimental results on segmentation of several dot patterns are presented.