Not exactly a solution, but maybe DBSCAN is an interesting clustering alternative, depending on your use case.