An algorithm for determining pattern connectivity and cluster compactness

A cluster-seeking algorithm incorporating graph-theoretical techniques is proposed. First, a totally connected weighted graph representing the set of patterns under consideration is formed. Second, those edges with weight values greater than a prespecified threshold T are removed from the graph. Next, the disjoint subgraphs corresponding to clusters are identified. Finally, the connectivity strength (or weakness) of each pattern is determined for detecting further separability of clusters. It is concluded that the proposed algorithm can be used in a computer-vision-related application where the objects of a scene are represented by clusters. The patterns can be characterized by the horizontal and vertical position of pixels and the corresponding depth (distance from camera). After applying the algorithm, the clusters are identified as the different objects in the scene