Abstract
The paper focuses on development of classification rules for road extraction from very high resolution satellite images. From the methodological point of view, a main emphasis is on object based image analyses and finding suitable features for discriminating consolidated (asphalt) roads from other land cover classes.
Results of practical tests on QuickBird images from the surroundings of Prague (combination of agriculture, urban and forest areas) are presented. Supervised, per-pixel approach was also applied.
A comparison of automatically derived land cover classes with manual interpretation of imagery showed similar level of accuracy of pixel and object based classification results. Nevertheless, a visual inspection proved better consistency of OBIA derived road segments that should be an input for creating a vector road network GIS layer.