Detection Of Urban Areas By Interference Filtering

Abstract

A new approach for detection of urban areas in satellite imagery with a resolution of 10 - 20 meters is proposed. The method, named interference filtering, is based on extraction of the road network from the imagery followed by a filtering operation that extracts areas where the line density is high. The resulting image is postprocessed by higher level algorithms to make a proposal for areas. The image can be combined with the old unrevised map of urban areas to make a proposal for a new revised map. Experiments indicate that interference filtering for detection of urban areas in satellite imagery is very promising. Most of the areas were detected. The border areas are, however, somewhat more uncertain. Interference filtering was superior to multispectral classification, even if texture features were included.