Voronoi-clustering for plane data

This paper presents a clustering algorithm based on Voronoi diagrams. The algorithm firstly constructs irregular grids in plane by Voronoi diagrams, then assign the points among different grids to different clusters according to the property of the Voronoi diagrams' “the nearest neighbor”. It is able to automatically modify the final clustering number based on the grid points' density, and it can adjust the locations for the Voronoi's seeds by the changes of the centroids, and the final Voronoi cells becomes the clustering result. The algorithm is able to settle down the clustering numbers automatically and also can recognize the low density points automatically. The experiments prove that the algorithm can cluster effectively the data points in plane, and its performance is similar to the X-means algorithm which is improved on the K-means algorithm. It is more effective than the DBSCAN and the OPTICS which are density-based clustering algorithms. The algorithm proved to be obviously more effective while the experimental data is in a larger scale.