ROBUST CONCURRENT APPROACH FOR ROAD EXTRACTION AND URBANIZATION MONITORING BASED ON SUPERPIXELS ACQUIRED FROM SPECTRAL REMOTE SENSING IMAGES

摘要

The extraction of road signatures from remote sensingimages as a promising indicator for urbanization is aclassical segmentation problem. However, somesegmentation algorithms often lead to non-sufficientresults. One way to overcome this problem is the usageof superpixels, that represent a locally coherent clusterof connected pixels. Superpixels allow flexible, highlyadaptive segmentation approaches due to the possibilityof merging as well as splitting and form new basicimage entities. On the other hand, superpixels require anappropriate representation containing all relevantinformation about topology and geometry to maximizetheir advantages.In this work, we present a combined geometric andtopological representation based on a special graphrepresentation, the so-called RS-graph. Moreover, wepresent the use of the RS-graph by means of a casestudy: the extraction of partially occluded road networksin rural areas from open source (spectral) remotesensing images by tracking. In addition, multiprocessingand GPU-based parallelization is used to speed up theconstruction of the representation and the application.