Incremental Delaunay triangulation construction for clustering

Octavio Razafindramanana (1) , Frédéric Rayar (1) , Gilles Venturini (2)

1 LIFAT - Laboratoire d'Informatique Fondamentale et Appliquée de Tours
2 Laboratoire d'Informatique

In this paper, we propose an original solution tothe problem of point cloud clustering. The proposed techniqueis based on a d-dimensional formulated Delaunay Triangulation(DT) construction algorithm and adapts it to the problem ofcluster detection. The introduced algorithm allows this detectionas along with the DT construction. Precisely, a criterion thatdetects the occurrences of gaps in the simplices perimeterdistribution within the triangulation is added during incrementalDT construction. This allows then to detect and label simplices asinter- or intra cluster. Experimental results on 2D shape datasetsare presented and discussed in terms of cluster detection andtopological relationship preservation.

Mots clés: Computational geometry; clustering; Delaunay triangulation; forward approach

Domaines: Géométrie algorithmique [cs.CG]; Apprentissage [cs.LG]

Octavio Razafindramanana, Frédéric Rayar, Gilles Venturini. Incremental Delaunay triangulation construction for clustering. 22nd International Conference on Pattern Recognition, Aug 2014, Stockholm, Sweden. ⟨hal-01094436⟩