We present a new variational model for saliency detection in images and its application to brain tumor segmentation. Incorporating a saliency term to a classical Total Variation based restoration functional this model is able to discriminate what is relevant (salient) from the background, resulting into a non-convex and non-smooth problem. To optimize the parameters of the proposed energy functional we introduce a Deep Learning framework using available knowledge from the specific application.

This presentation is part of Minisymposium “MS5 - Learning and adaptive approaches in image processing (2 parts)”
organized by: Kostas Papafitsoros (Weierstrass Institute Berlin) , Michael Hintermüller (Humboldt University and Weierstrass Institute Berlin) .

Authors:

Adrián Martín (Universitat Pompeu Fabra)

Emanuele Schiavi (Universidad Rey Juan Carlos)

Ivan Ramirez (Universidad Rey Juan Carlos)

Keywords:

deep learning, image segmentation, machine learning, nonconvex problems, nonlinear optimization, parameter learning, partial differential equation models