Comparison of Road Extraction in Urban Areas from High Resolution TerraSAR-X and IKONOS Images using Texture Features in Neural Network Algorithms

New generation of space radar sensors with high spatial resolution, heralds the possibility of these images usage for automatic feature extraction, specially road ones related. This study compares road extraction in high resolution radar and optical images. IKONOS and TerraSAR-X images are used for this purpose. In order to make possible the comparison, the images have been coregistered. Then, texture features have been extracted and classified using a back propagation neural network. By comparing the results with reference road network, provided by the operator, for TerraSAR-X and IKONOS images respectively, 46.10 and 57.72 percent for RCC, 46.58 and 93.27 percent for BCC and 0.61 and 0.31 for RMSE were obtained. Comparison of the outputs of two algorithms showed that each of optics and radar images have some defects in road extraction. For example, optical algorithms are sensitive to image parts which have the same texture and spectral features similar to roads such as parkings and roofs. In one hand, these areas have visible appearance and different texture in radar images, so radar images in urban areas especially with narrow roads and alleys, are suitable. On the other hand in areas with dense vegetation, Radar images do not perform well, while the optical images are able to distinguish the mentioned areas. In order to eliminate defects and increase the accuracy of the present algorithm, because of high functionality of these images, fusion of them seems to be an efficient wayin developing algorithms.