Like-feature detection in geospatial sources

The emergence of a new generation of satellites, increased dependence on computer-aided cartography, and conversion of paper-based maps along with the universal acceptance of the World Wide Web as a distribution medium, has resulted in widespread availability of geospatial data. Geospatial information systems have the potential to use this wealth of data to provide high-level decision support in important military, agricultural, urban planning, transportation and environmental monitoring applications. There are many challenges to take full advantage of this geo-spatial data collection. The first step in integration is to determine the correspondence between features in different sources. This problem, called like-feature detection is addressed in this paper. In addition to using the individual attributes of features, we use the geographic context abstracted as proximity graphs, to improve the matching process. The proximity graph models the surroundings of a feature in a source and provides a measure of similarity between features in two sources. Pair-wise similarity between features of two sources is then extended to multiple sources in a graph-theoretic framework. Experiments conducted to demonstrate the viability of our approach using a variety of data sources including satellite imagery, maps, and gazetteers show that the approach is effective.