Research On Road Extraction Methods From Remote Sensing Images

Road network is very important geographic information, with the rapid development of information technology in recent years, high-resolution remote sensing satellites provide mankind with high-definition and large-capacity remote sensing satellite imagery. At the same time people propose many automatic and semi-automatic road extraction methods and use these methods to abstract spatial information from high resolution images. This paper mainly studies the template matching method, mathematical morphology method combined with hough transform, region growing method and the method based on snakes model, and use these methods to extract road from remote sensing images.This paper makes some improvement as follow:(1) Improve the degree of similarity criteria in template matching method, the improved similarity criteria is: get the absolute difference between initial template window's pixel gray value and target window's pixel gray value, make the squares of absolute difference's sum less than the set threshold. The experimental results verify the improved method can effectively improve the accuracy of the road extraction.(2) Improve the establishment of initial template window, the establish initial template window perpendicular to the road direction ; set the rule of center deviate from centerline. The initial template window perpendicular to the road direction, makes the search along the road direction and ensure the correctness of the search direction. Set the rule of center deviate from centerline can guarantee the detected center in centerline, reduce errors and improve accuracy of detection.(3) In the method based on snakes model, optimize and improve the parameters of snakes algorithm and make corresponding simulation. The experimentation achieved good results.