Prof. Klas Modin, Chalmers University of Technology, Gothenburg,
Diffeomorphic Image Registration Based on Fisher Information
In this talk I will explain a new geometric method for image registration,
based on a recently discovered connection between information geometry
and topological hydrodynamics. Images are thought of as probability
densities. The objective is to find a diffeomorphism that warps a
source density so that its gross features match a target density.
The diffeomorphism should be optimal with respect to an energy functional
derived from the Fisher--Rao metric. We obtain the solution by a gradient
flow on the space of diffeomorphisms. This flow admits a geometric
reduction to a 2-component density gradient flow. Connections to optimal
mass transport will be discussed.
October 17, 2014
Location: Bahen 6183
Nir Sochen, Tel-Aviv University
Certainties and Uncertainties in the Principle of Uncertainty
Measurement of features is important in signal/image processing and
in physics and other branches of science and technology as well. The
principle of uncertainty enforces limitations on the accuracy of the
measurement of joint features. We show how these limitations can be
overcomed in most cases and give a new interpretation to this principle
in a way that leads to interesting generalisations.
Joint work with H-G. Stark and R. Levie