OBJECT-ORIENTED ROAD EXTRACTION BASED ON IMPROVED FCM AND SHAPE FILTER IN HIGH RESOLUTION RS IMAGERY

Abstract

An object-oriented method of road extraction is proposed for high-resolution remote sensing imagery aiming to its characteristics.  At first, bilateral filter is used in original imagery to smooth detail information and retain road edge; then it proposes an improved Fuzzy C-Means algorithm combined with the neighborhood information in order to deal with outliers better, so the imagery is segmented to independent objects by improved FCM algorithm; and then it filters every objects by geometric feature, after that it connects road segments to get network by region growing algorithm and executes post-processes by morphology method.  The experiments show that the method can extract the road target efficiently from high resolution imagery with higher accuracy.