Exploring an EM interpretation of the Softassign algorithm for alignment problems

John T. Kent, Kanti V. Mardia & Charles C. Taylor

The alignment problem for two unlabelled configurations of points involves finding a suitable subset from each configuration such that one subset can be nearly superimposed on the another, up to a rigid body transformation. Rangarajan et al. (1997) formulated the ``Softassign'' algorithm to tackle this problem. In this paper we show that their algorithm can also be interpreted as an EM algorithm.