A Novel Information-Theoretic Point-Set Measure Based on the Jensen-Havrda-Charvat-Tsallis Divergence

A novel point-set registration algorithm was proposed in [6] based on minimization of the Jensen-Shannon divergence. In this contribution, we generalize this Jensen-Shannon divergence point-set measure framework to the Jensen-Havrda-Charvat-Tsallis divergence. This generalization permits a fine-tuning of the actual divergence measure between robustness and specificity. The principle contribution of this submission is theitk::JensenHavrdaCharvatTsallisPointSetMetric class which is derived from the existing itk::PointSetToPointSetMetric. In addition, we provide other classes with utility that would extend beyond the point-set measure framework that we provide in this paper. This includes a point-set analogue of the itk::ImageFunction, i.e. itk::PointSetFunction. From this class we derive the class itk::ManifoldParzenWindowsPointSetFunction which provides a Parzen windowing scheme for learning the local structure of point-sets. Finally, we include the itk::DecomposeTensorFunction class which wraps the different vnl matrix decomposition schemes for easy use within ITK.