Doctorant sous la direction de F. Pascal

Gordana Drašković
Ph.D Student
CentraleSupélec, Gif-sur-Yvette
Laboratory of Signals and Systems (L2S), Gif-sur-Yvette
Signals and Statistics Group – Modelisation and Estimation Team

L2S/CentraleSupélec
gordana.draskovic@l2s.centralesupelec.fr

Thesis Director: Frédéric Pascal

Co-director: Florence Tupin

Thesis title: Robust Estimation Analysis for Image Processing

Started in October 2016.

Thesis abstract: Nowadays, one of the most challenging problems is to extract and analyze information of interest from high-dimensional observations and possibly heterogeneous, containing outliers or missing data. To tackle this problem, robust estimation theory offers many possibilities, notably the robustness to heavy-tailed distributions and to outliers. Nevertheless, its adaptation for learning in large-scale data context requires more in-depth research. For instance, one can mention the performance characterization in large dimensional context, the existence of estimators, the computational cost... The aim of this work is first to characterize and quantify the contribution of these estimators for clustering purposes, images indexation, detection with application to sequences of remote sensing images. Finally, the results of this work, theoretical and applied, will give rise to publications in the best journals and conferences of the field. In addition, this work will contribute to the animation of the PASADENA WG through presentations and discussions with members of this WG on this research topic.