A probabilistic approach to spectral graph matching
Yosi Keller and Amir Agozi
IEEE Transactions on Pattern Analysis and Machine Intelligence 2009.

Abstract

Spectral matching is a computationally efficient approach to approximately solving pairwise matching problems that are \textit{np}-complete. In this work we present a probabilistic interpretation of spectral matching schemes and derive a novel probabilistic matching scheme that is shown to outperform previous approaches. We show that spectral matching can be interpreted as a maximum likelihood estimate of the assignment probabilities and that the Graduated Assignment algorithm boils down to an Estimate Maximize formulation. Based on this analysis we propose a novel iterative probabilistic matching scheme that relaxes some of the inherent assumption used in prior works and efficiently handles erroneous inputs. We show our approach to outperforms previous schemes when applied to exhaustive synthetic tests as well as the analysis of real image sequences.

[Edit]