Distance measures between attributed graphs and second-order random graphs

Abstract

The aim of this article is to purpose a distance measure between Attributed Graphs (AGs) and Second-Order Random Graphs (SORGs) for recognition and classification proposes. The basic feature of SORGs is that they include both marginal probability functions and joint probability functions of graph elements (vertices or arcs). This allows a more precise description of both the structural and semantic information contents in a set (or cluster) of AGs and, consequently, an expected improvement in graph matching and object recognition. The distance measure is derived from the probability of instantiating a SORG into an AG. SORGs are shown to improve the performance of other random graph models such as FORGs and FDGs and also the direct AG-to-AG matching in two experimental recognition tasks.