java - DBSCAN libraries to extract density-reachable points -

i'm working dbscan libraries extract clusters set of data. far i've tested dbscan using apache common math , weka libraries. (my question not libraries available implementations of dbscan)

so far i've understood in dbscan there 3 types of points (as according wikipedia): core points, (density-)reachable points , outliers. issue need extract clusters , it's frontier points or density-reachable points.

do know dbscan library allows me extract density-reachable points per cluster?

in elki implementation, can use options

-algorithm clustering.gdbscan.generalizeddbscan -gdbscan.core-model 

to cluster "model" containing core points of cluster only. cluster members still border points - density reachable, not core. however, needs more memory, not enabled default.

in image, inner convex hull core points only. green cluster, there 2 core points. noise points, there no nested cluster, obviously.

note dbscan clusters can non-convex. why green cluster can have core points inside convex hull of red cluster. not every point inside inner hull is core point. there noise point right inside of red cluster, , not error - data set sparse, has local density variations epsilon , minpts. point in vincinity of noise point cannot core point; point of inner convex hull 1 sure.

the cluster objects provide full list of points, not convex hull. core points accessible via clusters coreobjectsmodel. visualization code uses convex hulls avoid cluttering image much. also, default output writer not output information. need use java, , either write custom resulthandler output data desired, or in elki.

note distinction between border points, noise points , core points considered obsolete , not supported theoretical models in newer literature.