Abstract

Geometric Methods in 2D and 3D image matching

Martin Rumpf
University of Duisburg, Germany

Image registration and image matching are classical problems in computer vision.On the one hand abstract gradient flows on manlfolds allow a new perspective to the regularization of the problem. Here the manifold is the space of deformations $\phi$ , where the metric $g$ on this space is supposed to be appropriately regular and may depend on the
images $T$, $R$ to be mapped and the deformation $\phi$ itself.
On the other hand one can consider an energy regularization and
ask for morphological matching methods.
In an geometric setting this problem can be rephrased
as the matching problem for the Gauss maps of the images.