CMIC Seminar: Continuous Multi-Label Optimization and Diffusion MRI

Speaker: Mohamed Souiai, Technical University of Munich
UCL Contact: Dominique Drai (Visitors from outside UCL please email in advance).
Date/Time: 11 May 16, 13:00 - 14:00
Venue: Roberts 106
Further Information:

Light refreshments provided.

Abstract

In this talk I will present my recent works which arrise from applications such as in semantic image segmentation and joint motion segmentation of RGB-D images. These labelling problems can be cast as variational problems in which we can incorporate so called co-occurrence, hierarchical, proximity and label transition priors in order to improve the resulting labelling.

Additionally, I will introduce a novel approach to improving the integrality of the solution to relaxed multi-labelling problems. To this end, we incorporate the entropy of the objective variable as a measure of the relaxation tightness.

Finally, I will also present novel methods for diffusion MRI processing which allow 12-fold shorter scans and model-free disease detection.