Comparison of spatial point patterns and processes characterization methods
Laboratoire de Cancérologie Expérimentale, CJF INSERM 9311, IFR Jean
Roche, Faculté de Médecine Nord - Boulevard Pierre Dramard, 13916 Marseille
Cedex 20, France
Accepted: 17 April 1998
The topographical analysis of spatial point patterns is a way to quantitatively characterize the organization of those patterns in either computer-based (percolation, cellular automata, ) or experimental (thin films, alloys, cell biology, astronomy) models. We have tested the five most used methods (nearest-neighbour distribution, radial distribution, Voronoï paving, quadrat count, minimal spanning tree graph) which generate nine parameters on stochastic models (random point process, hard disks model and cluster models) and locally perturbed lattices models. The methods of topographical analysis were compared in terms of discriminant power, sensitivity to local order perturbations, stability of parameters, methodological bias and algorithmic. The method which offers the best discrimination power and stability appears to be the minimal spanning tree graph edge length distribution.
PACS: 06.30.-k – Measurements common to several branches of physics and astronomy / 42.30.Sy – Pattern recognition / 02.50.Ey – Stochastic processes
© EDP Sciences, 1998