Florence Tupin

A probabilistic framework for patch-based estimation of noisy data and its application to SAR imagery


In this presentation, a probabilistic framework for patch-based non-local approaches will be described and applications in SAR imagery presented. First, an introduction to SAR imagery, its specificities and its different modalities (amplitude, interferometry, polarimetry), will be given. The statistical models existing for these data, which are highly noisy due to the speckle phenomenon, will be presented. Then, we will describe how patch-based non local approaches can be formulated in a probabilitic framework. In this case, the problem is seen as a weighted maximum likelihood estimation problem, and allows to process any data for which a distribution model of the noise is available. In the case of SAR imagery, a unified framework can be successfully applied to amplitude images, interferometric data and polarimetric data. This work has been done in collaboration with Charles Deledalle and Loïc Denis.