A comparison of Junction Tree and Relaxation Algorithms for point matching using different distance metrics

We have developed a polynomial time optimal method for a class of attributed graph matching problems using the Junction Tree algorithm from Graphical Models. In this paper we compare this method with standard probabilistic relaxation labeling using different forms of point metrics and under different levels of additive noise. Results show that, no matter which of the metrics is applied, our technique is more effective than probabilistic relaxation labeling for large graph sizes. For small graph sizes, our technique is still preferable for two of the metrics, while for the third one both techniques perform similarly.