Transport optimal pour assimilation de données images

English

Séminaire Doctorants

20/10/2015 - 15:00 Mr Nelson FEYEUX Salle 1 - Tour IRMA

Meteorology is devoted to forecasting future weather. Complex models are used to compute the future states of the atmosphere and the oceans, given an initial condition (the current state of the system). This initial condition cannot be measured directly. Instead, previous forecasts as well as recent measures and satellite images are combined to guess this current state of the system. This is called data assimilation.

We propose to use optimal transport techniques in order to assimilate satellite images. Optimal transport is a mathematical theory which, due to some of its interesting properties, is widely applied to image processing. The Wasserstein distance is used to write a cost function that has to be minimized. Some results are shown and underlying problems are discussed.