Nondeterministic kinetics based feature clustering for fractal pattern coding

Newton potential is reformulated in terms of the Hausdorff distance to design reduced affine mappings associated fractal attractors. By applying maximum entropy analysis to observed patterns, stochastic features are extracted as well as boundary points where the fixed points of the mappings should be located. To linear segments of potential fixed points, feature points are nondeterministically attracted following the Hausdorff potential. Guided by this feature clusters, random patterns are partitioned to estimate mapping parameter.