Spectral Descriptors for Graph Matching

04/04/2013

∙

by   Nan Hu, et al.

∙

0

∙

share

In this paper, we consider the weighted graph matching problem. Recently, approaches to this problem based on spectral methods have gained significant attention. We propose two graph spectral descriptors based on the graph Laplacian, namely a Laplacian family signature (LFS) on nodes, and a pairwise heat kernel distance on edges. We show the stability of both our descriptors under small perturbation of edges and nodes. In addition, we show that our pairwise heat kernel distance is a noise-tolerant approximation of the classical adjacency matrix-based second order compatibility function. These nice properties suggest a descriptor-based matching scheme, for which we set up an integer quadratic problem (IQP) and apply an approximate solver to find a near optimal solution. We have tested our matching method on a set of randomly generated graphs, the widely-used CMU house sequence and a set of real images. These experiments show the superior performance of our selected node signatures and edge descriptors for graph matching, as compared with other existing signature-based matchings and adjacency matrix-based matchings.