Abstract: The need for diffeomorphic matching of unlabeled points arises in numerous registration and shape analysis applications in biomedical imaging. Solving for a diffeomorphic transformation to interpolate two sets of points with outliers and unknown correspondence is a challenging problem. In this work, we use a strategy of joint clustering to find the correspondence at the cluster level and a diffeomorphic transformation in space. The cluster centers are pre-labeled. As the cluster centers evolve during the iterations of a deterministic annealing EM algorithm, we estimate a diffeomorphism in space. The cluster centers with the same index in the two point sets are always in correspondence. The number of clusters can be estimated by minimizing the Jensen-Shannon divergence on the point data. The transformation is symmetric with respect to inversion of the two point sets being matched. We show application of the algorithm on image registration on corpus callosum and hippocampus shapes.
This is joint work with Anand Rangarajan, Department of Computer and Information Science and Engineering, University of Florida.
This seminar is easily accessible to persons with disabilities. For more information or for assistance, please contact the Mathematics Department at 743-3500.