Urban Road Network Extraction in SAR Images Exploiting Road Junction Knowledge

Abstract:
In this paper we propose a new method for road extraction in SAR images designed to be fully automatic and aimed at optimal junction detection. The algorithm exploits spatial relationships between a pixel and its neighbors, looking for straight features in the image and combining multiple detectors to improve road candidate extraction. The second part of the procedure tries to reconstruct the road network selecting among the segment candidates those minimizing a weight functional. The result is obtained by defining a Markov Random Field (MRF) on the set of segments just gained. However, in addition to introducing contextual knowledge about the shape of road objects, suitable “hint elements” to the Markovian framework based on road junctions are added. The last step of the procedure is based on perceptual grouping rules, applied in order to improve the reconstruction results by eliminating redundant lines and providing a better alignment of road elements in the final map.