Algorithmic Solutions for Classes of Labeling Problems

                          Torsten Wierschin
                 Institute for Artificial Intelligence
                        Faculty of Informatics
                   Dresden University of Technology


The graph-based approach is now an established part in the works of
the PR/CV communities, see e.g.
http://www.iapr-tc15.unisa.it/TC15b/TC15.html.

We are concerned with tasks which can be stated within the statistical
framework of labeling problems, a formalism related to structural
properties of graphs, like cliques or proper colorings together with
an appropriate probability distribution defined on them.

A summary of this area will be given with its application to image
segmentation, registration and dump-CAD-data interpretation.

We report on achievings and ongoing studies using techniques of
randomisation, approximation and sampling.