Graph matching and clustering using spectral partitions

作者:

Highlights:

摘要

Although inexact graph-matching is a problem of potentially exponential complexity, the problem may be simplified by decomposing the graphs to be matched into smaller subgraphs. If this is done, then the process may cast into a hierarchical framework and hence rendered suitable for parallel computation. In this paper we describe a spectral method which can be used to partition graphs into non-overlapping subgraphs. In particular, we demonstrate how the Fiedler-vector of the Laplacian matrix can be used to decompose graphs into non-overlapping neighbourhoods that can be used for the purposes of both matching and clustering.

论文关键词:Inexact graph matching,Graph simplification,Graph partition,Graph spectral clustering,Fiedler vector

论文评审过程:Received 20 December 2004, Revised 30 June 2005, Accepted 30 June 2005, Available online 13 September 2005.

论文官网地址:https://doi.org/10.1016/j.patcog.2005.06.014