Road Extraction from Satellite Images using a Fuzzy-Snake Model

This paper proposes a developed approach to extract roads from optical remotely sensed images. The approach is based on the following steps. First, a window with size of 565 pixels is moved over the image to calculate the features: mean (x(1)), standard deviation (x(2)), skewness (x(3)) and kurtosis (x(4)). Then, the roads are identified based on the converted features to the specific fuzzy sets of the linguistic variables. The used linguistic variables are Mean, Standard deviation, Skewness, Kurtosis and Grey-scale with trapezoid and triangle membership functions. Next, the skeleton of the identified roads is extracted using two structure elements from the mathematical morphology. Finally, a snake model is employed to extract the road vector form from the skeletons. The results of the accuracy evaluation demonstrate that the developed road extraction approach can provide both good visual and high positional accuracy. The approach is tested over the samples of SPOT-4 panchromatic images from areas in Iran.