A Study Of Road Extraction Based On Steerable Filter

With the developing of remote sensing, we can get more and more information from the satellite. There are many difficulties which are how to deal with and analyze these images. Road information is important geographical special information. Road network extracted from remote sensing image is used in military, survey mapping, traffic, navigation, etc. Although technology of road extraction has made a excited progress from 1995, most of good methods need manual help. It is a problem to how to find an automatic and effective method for extracting road information from satellite images.The paper develops the theory of road extraction based on data of high resolution remote sensing image. Above all, three steps of road extraction are described in this paper. They are image enhancement, edge detection and edge linking. Then advantages and disadvantages of classical methods in every step are analyzed and evaluated. At last, road directions are used in edge detection and edge linking. Edge enhancement and edge detection based on steerable filter and edge linking which is carried out by utilizing the direction information are used in the experiment of high resolution remote sensing image. The results proves that the road extraction based on steerable filter can detect the edge of road automatically and obtain the width of road exactly.Image enhancement is used to enlarge the difference between edge information and background. Three good methods are presented, which are Histogram Equalization, Homomorphic Enhancement and Multi-Resolution Wavelet Enhancement. Their theories are introduced in detail. The enhanced image using Histogram Equalization clearly displays that the edge information included in the image has been enhanced.Edge detection is a very important step in road extraction. It is a hard work for road extracting automatically to finding more continuous edges of road, at the same time, depressing the other edges information. As we all know that the method of Canny edge detection is the best one in the world. The theory of Canny edge detection is analyzed in the paper. We put forward combining the Canny edge detection with the steerable filter to improve the method of Canny. The new method includes the acuity of Canny detecting edge and the flexibility of steerable filter for road direction. The paper proves that statistic direction of road is feasible. The statistic direction of road in the image is used to direct the steerable filter filtering the image, and the result proves that steerable filter is valid in enhancing and extracting the edges of road.Edge linking is to select and link the discontinuous edges in the base of the result of edge detection. Hough Transformation is used in this paper to hold the edges of road and delete the other edges information. We improved the Hough Transformation and make it searching edges only along road direction, and the experiment proved it is a good way. The Visual Perceptual Edge Linking based on Two-Level Threshold and the Novel Heuristic Search Algorithm are all applied to link the edges of road synchronously. The last result of road extraction proves thatthese new method are valid and the edges of road are extracted exactly and completely.