Extraction of Road Pixels from Airborne Image and Elevation Data: Focusing on Feature Selection

摘要: Road pixel segmentation in airborne data is an important and challenging task. Recently, a sophisticated robust approach based on superpixels minimum cost paths has been published. In order to find out which of the numerous features are most essential, we propose forward-search wrapper for feature selection was tested with two different classifiers both generic customized features. Path connecting high probability being roads then established, filtered, included as higher cliques into non-local energy minimization module thus enforcing connectivity resulting road network. Two minor contributions adjustment algorithm multichannel images efficient application Dijkstra path method sparse set start end nodes. The results show that yield quite sets, speaks favor approaches. Also, were ranked among relevant, emphasizing importance sensor fusion higher-level semantical concepts.