Graph Methods for Discrete Signal Processing

About

In the recent years, graph-based methods have become increasingly popular in the mathematical community due to their flexibility when applied to large data clustering/segmentation problems. The main idea of these models consists of building a graph from the given data by means of a nonlocal distance measure with respect to suitable features of the given data.

Many classical variational and PDE models defined in a continuum setting can be translated and reinterpreted in this discrete graph framework. Applying suitable optimisation strategies are needed to reduce the computational costs due to the large size of the data considered. In this session we gather researchers in the field of graph models in order to discuss some recent developments of this framework, focusing in particular on its applications to image, 3D surface and higher dimensional point cloud processing.

Course leader

François Lozes

Target group

Phd students

Fee info