This presentation give an overview of the curvilinear object extraction framework described by Thierry Geraud. This framework relies on four main steps : removing image minima using an area closing, computing the watershed transform of the filtered image, building a "curve adjacency graph" and defining a Markovian Random Field to select the interesting nodes of the graph.

This framework can be easily adapted for a lot of applications. We use the extraction of road networks as an example during this presentation.

We begin by describing each steps and explaining why each one is needed. We then show how to adapt the framework to different problems and its drawbacks. Finally we give some ideas and tests results about improving the method.

• curvilinear_object_extraction.pdf