Many-to-many feature matching under infinity norm distance for object recognition

Graph matching is an important component for many object recognition algorithms. Most of the graph matching algorithms search for one-to-one correspondence between graph vertices. Yet it is more realistic to compare a vertex subset of a graph to a vertex subset of one other graph in a many-to-many fashion. In this thesis, a similarity relation between graph vertices are presented by many-to-many matching approach. To conduct many-to-many matching on graphs, graphs are represented as point sets in normed spaces. There are some useful algorithms to compare point sets that work in polynomial time and permit many-to-many matching in a natural way. Consequently, graph matching problem is stated as that of point matching in this thesis. One of our primary objectives is to maintain isometric embedding of graphs into normed spaces. According to the results of the experiments, matching by reference points technique is developed to increase the efficiency of this approach. The results that are obtained in this manner are explained and the experimental evaluation of the algorithm is demonstrated.