Matching attributed graphs: 2nd-order probabilities for pruning the search tree

Abstract

A branch-and-bound algorithm for matching Attributed Graphs (AGs) with Second-Order Random Graphs (SORGs) is presented. We show that the search space explored by this algorithm is drastically reduced by using the information of the 2nd-order joint probabilities of vertices of the SORGs. A SORG is a model graph, described elsewhere, that contains 1st and 2nd-order order probabilities of attribute relations between elements for representing a set of AGs compactly. In this work, we have applied SORGs and the reported algorithm to the recognition of real-life objects on images and the results show that the use of 2nd-order relations between vertices is not only useful to decrease the run time but also to increase the correct classification ratio.