Graph Matching via Sequential Monte Carlo

Yumin Suh, Minsu Cho, Kyoung Mu Lee

Abstract: Graph matching is a powerful tool for computer vision andmachine learning. In this paper, a novel approach to graph matchingis developed based on the sequential Monte Carlo framework. By con-structing a sequence of intermediate target distributions, the proposedalgorithm sequentially performs a sampling and importance resamplingto maximize the graph matching objective. Through the sequential sam-pling procedure, the algorithm effectively collects potential matches un-der one-to-one matching constraints to avoid the adverse effect of outliersand deformation. Experimental evaluations on synthetic graphs and realimages demonstrate its higher robustness to deformation and outliers.