Link and bottleneck link set functions and multiresolution structural clusters

Abstract

A method for structural clustering is proposed based on information about entity-to-set interactions that can be easily collected in digitalized images. The method is based on the multiscale structure of the problem of maximization of a so-called bottleneck link function, which can be globally solved with a greedy-wise algorithm. Extension of the construction to clusters consisting of "hard" cores and dispersed "shells" is proposed.