Level Sets and Voronoi based Feature Extraction from any Imagery

Polygon features are of interest in many GEOProcessing applications like shoreline mapping, boundary delineation, change detection, etc. This paper presents a unique new GPU-based methodology to automate feature extraction combining level sets, or mean shift based segmentation together with Voronoi skeletonization, that guarantees the extracted features to be topologically correct.

The features thus extracted as object centerlines can be stored as vector maps in a Geographic Information System after labeling and editing. We show application examples on different sources: paper maps, digital satellite imagery, and 2D/3D acoustic images (from hydrographic surveys). The application involving satellite imagery shown in this paper is coastline detection, but the methodology can be easily applied to feature extraction on any king of imagery.

A prototype application that is developed as part of this research work.