Automatic Road Extraction from Remote Sensing Imagery Incorporating Prior Information and Colour Segmentation

In this paper an approach to road extraction in open landscape regions from IKONOS multispectral imagery is presented which combines a line-based approach for road extraction with area-based colour segmentation. Existing road databases are used in two ways: firstly, to estimate scene dependent parameters of the line-based approach and secondly to exclude non road regions from the extraction, also exploiting the colour information. The images and reference data for the evaluation are identical to the information used in a recently conducted EuroSDR test on automatic road extraction. Therefore, the results can easily be compared with the published ones. It is shown that our new approach with reduced human interaction for the parameter optimisation obtains results similar to the best ones of the EuroSDR test.