Adaptive directional order filters and mathematical morphology for road network extraction on SAR images

In this paper, a method for automatic detection of linear features in SAR images, with application to road network extraction is proposed. A pre-filtering step is firstly performed using an adaptive directional weighted order filter. It aims to smooth the noise, enhance edges and preserve thin anisotropie structures. Then, the detection operator is applied : it is based on a geometrical model of the roads and uses morphological directional operators with appropriate structuring elements. Results are presented on real SAR ERS-1 satellite data.