Incorporating Generic and Specific Prior Knowledge in a Multiscale Phase Field Model for Road Extraction From VHR Images

Abstract

This paper addresses the problem of updating digital road maps in dense urban areas by extracting the main road network from very high resolution (VHR) satellite images. Building on the work of Rochery <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">et</i> <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">al.</i> (2005), we represent the road region as a ldquophase fieldrdquo. In order to overcome the difficulties due to the complexity of the information contained in VHR images, we propose a multiscale statistical data model. It enables the integration of segmentation results from coarse resolution, which furnishes a simplified representation of the data, and fine resolution, which provides accurate details. Moreover, an outdated GIS digital map is introduced into the model, providing <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">specific</i> prior knowledge of the road network. This new term balances the effect of the <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">generic</i> prior knowledge describing the geometric shape of road networks (i.e., elongated and of low-curvature) carried by a ldquophase field higher order active contourrdquo term. Promising results on QuickBird panchromatic images and comparisons with several other methods demonstrate the effectiveness of our approach.