|Computing the Earth Mover's Distance under Transformations|
This project aims at extending the Earth Mover's Distance (EMD) between distributions to be invariant to some given set of distribution transformations. This problem is motivated by examples from content-based image retrieval. A monotonically convergent iteration is given, although the iteration may converge to only a locally optimal transformation. Specific cases are identified in which a globally optimal transformation can be computed directly, without the aid of our iteration. A gentle introduction to the EMD is also included.