Fachvortrag in englischer Sprache

Osmosis is omnipresent in biological systems. It describes a conservative transport process with nonconstant equilibrium. In this talk we show how related ideas can be used to solve image processing and computer graphics problems. The simplest filters with an osmosis-like behaviour can be modelled by a linear drift-diffusion process, also known as Fokker-Planck equation or Kolmogorov forward equation. In contrast to pure diffusion methods, the additional drift term allows to design filters with interesting steady states. We show how this steady state can be controlled by the drift vector field, and we interpret osmosis models as mechanisms for data integration. We discuss important theoretical properties of osmosis filtering, and we show its usefulness for a number of applications.

Speaker: Prof. Dr. Joachim Weickert, Saarland University