SITIS'06 Tutorials

Parametric and Evolutionary Methods in Image Processing
Jean Louchet, Evelyne Lutton
INRIA, France

In the first part of this tutorial, we will show how the main current approaches to image analysis and segmentation may be revisited and understood as heuristics to efficiently explore parameter spaces. Extending this philosophy gives a new life to Hough-inspired methods: we show how it is possible to introduce evolutionary techniques into them in order to speed up and extend the applications of parametric image analysis.

The second part will concentrate on the Parisian approach, a class of evolutionary algorithms where the solution is represented by an evolving population rather than single individuals. Its main application, the Fly Algorithm, originally a fast stereovision method, is now applied to real-time robot vision and planning, requiring less computing power and giving faster reaction to external events than classical approaches. We will get into its programming and tune-up, and describe prospective applications to signal or medical image processing.

Tutorial slides are available here.