The Role of Graph Topology for Graph Matching

Graph matching plays a key role in structural pattern recognition. However, for concrete application, an appropriate feature graph topology must be chosen. In this paper, this issue is investigated by comparing the performance of four different graph topologies with respect to four types of feature graphs and two algebraically graph matching methods: least squares method (LSM) and eigenspectral methods. Results clearly demonstrate that standard delaunay triangulation topology is not as successful as other nearest neighbor models for graph matching.