Road Extraction In Remote Sensing Images Based On Image Features

In recent years, remote sensing technology has made rapid development. How to deal with remote sensing images effectively, and get information they need from them quickly and accurately, has becoming a very popular hotspot in the area of remote sensing research.While dealing with the remote sensing images, extracting the linear ground objects especially the roads plays a very basic and important part. This paper introduces the relative theories of image processing, analyzes and uses some of the existing road-extraction algorithms for reference, basing on which, author attempts to recognize ground objects and then extract roads by using computers to simulate the human's habit of feature recognition. The thesis presents a new algorithm of road extraction based on image features especially for the high-resolution colored remote sensing optical image. First of all, it filters pixels according to their color information, and then it uses the method of regional expansion with seeds, combining the fuzzy function to reduce interference from noise. It also filters objects in the image selectively based on area parameter by combining "Open" and "Close" morphological operations to get the roads needed at last. According to the experiments research on remote sensing images by using the algorithm above, the results shows that the algorithm presented in this paper could simulate the human's habit of feature recognition, and enhance the integrality of the extracted roads.Finally, this thesis takes out an interactive image processing system, which could process remote sensing images with functions of geometrical transformations, edge detection and smoothing filter. The system also offers the function of semiautomatic road-extraction from high-resolution colored remote sensing images conveniently and effectively, and shows references to the width and direction of the roads at the same time.