Automatic Road Extraction Using High Resolution Satellite Images

Nowadays analysis of the high resolution satellite images is becoming an important research subject for analysis of urban areas. Automatic road network extraction is an important aspect of the urban analysis in the urban areas. We proposed two approaches to extract the road, they are Level Set and Mean Shift methods.

Due to the presence of other road-like features with straight edges, extracting the roads from an original image is very difficult and computationally costly. We pre-processed the image and by reducing the noise (the buildings, parking lots, vegetation regions and other open spaces) we can improve the tolerance. Firstly, we will extract the roads as elongated regions. Based on the fact that road networks constitute large number of small linear structures, we can remove non-linear noise segments with the help of a median filter. Then by using Level Set and Mean Shift method, we can perform the road extraction.

Finally, based on quality measures we can evaluate the accuracy of the road extracted images. The 1m resolution IKONOS data can be used for the experiment.


We developed this project using VHSIC Hardware Description Language (VHDL).

VHSIC – Very High Speed Integrated Circuit



ModelSim XE III 6.4b: is used for Simulation.

Xilinx ISE 10.1: is used for Synthesis.