Urban road extraction from VHR images using a multiscale approach and a phase field model of network geometry

This paper addresses the problem of automatically extracting the main road network in a dense urban area from a very high resolution optical satellite image using a variational approach. The model energy has two parts: a phase field higher-order active contour energy that describes our prior knowledge of road network geometry, i.e. that it is composed of elongated structures with roughly parallel borders that meet at junctions; and a multi-scale statistical image model describing the image we expect to see given a road network. By minimizing the model energy, an estimate of the road network is obtained. Promising results on 0.6m QuickBird Panchromatic images are presented, and future improvements to the models are outlined. I.