Vous êtes ici

Séminaire du jeudi 18/10/2012

Date: 
Jeudi, Octobre 18, 2012 - 14:00 to 16:00
localisation: 
Bat A9, Salle: Amphi Kastler

--------------------------------------------------------------------------------

Orateur : Nicolas Papadakis

Affiliation : CNRS, LJK, Grenoble et IMB, Bordeaux

Titre : Convexification of multi-label problems for image processing

Résumé : Numerous image processing problems rely on the partitioning of an image in different classes. Such classes can be represented by labels belonging to a set of discrete possible values, that can possibly be non ordered. The minimization of non convex energies with respect to functions taking their values in such sets is nevertheless NP hard. For that reason, most popular methods are only able to compute local minima, using linearization, multi-resolution, graph representation... In this talk, recent works offering an interesting alternative will be presented. They are based on the convexification of non convex energies and are applied to a large range of classical problems: segmentation, optical flow computation, inpainting... With such reformulation of the problem, convex optimization tools can be used to obtain fast and accurate estimations of global minima of the original energies.

--------------------------------------------------------------------------------

Orateur : Pierrick Coupé

Affiliation : LaBRI

Titre : Scoring by Non-local Image Patch Estimator for Early Detection of Alzheimer’s Disease

Résumé : In this talk, I will present an innovative approach to robustly and accurately detect Alzheimer's disease (AD) based on the distinction of specific atrophic patterns of anatomical structures such as hippocampus (HC) and entorhinal cortex (EC). The proposed method simultaneously performs segmentation and grading of structures to efficiently capture the anatomical alterations caused by AD. Known as SNIPE (Scoring by Non-local Image Patch Estimator), the novel proposed grading measure is based on a nonlocal patch-based frame-work and estimates the similarity of the patch surrounding the voxel under study with all the patches present in different training populations. A validation of the ADNI database will be presented and a comparison with recent studies on MRI-based biomarkers will be detailed.

--------------------------------------------------------------------------------