A dynamic generating graphical model for point-sets matching

This paper presents a new dynamic generating graphical model for point-sets matching. The existing algorithms on graphical models proved to be quite robust to noise but are susceptible to the effect of outliers. We discuss the separator’s influences on point-sets matching in inference on graphical models theoretically and find that the separator which consists of outliers will break the message-passing, which will directly lead to the breaking down of the existing methods. Due to the conclusion above, in order to minimize the outliers in separator we propose a new algorithm in generating a graphical model and the corresponding Junction Tree for point-sets matching. The experimental results show that the proposed algorithm is significantly more stable and possesses higher accuracy on point-sets matching, which can overcome the limitation of sensitivity on outliers in the existing graphical models.