Introducing authority and hubness into graph matching

Abstract

Spectral matching is an efficient approach for inexact graph matching. Many spectral matching methods boil down to power iteration which calculate the confidence vector iteratively. Inspired by the Web page ranking method Hypertext Induced Topic Search (HITS), we introduce hubness vector and authority vector to replace the traditional confidence vector, and an iterative algorithm is proposed to solve the subgraph matching problem. The incorporation of hubness and authority can help reduce the distraction caused by outliers, and provides better robustness against outliers. The performance of the proposed algorithm is evaluated on both synthetic graphs and real-world images.