Research Of Road Extraction Algorithm From Remote Sensing Images Based On NSCT

With the development of the technology and wide application of various space sensors, information extraction from remote sensing images has become an important technique. How to get information from these remote sensing images with high speed and high precision is still a difficulty or focus in this field. The roads in the remote sensing images are the important got-up objects as well as city skeleton. Road extraction is very meaningful to the city programming, military affairs aim reconnaissance as well as traffic management and so on.Eliminating noises using any kind of image enhancement ways, then segmenting and extracting roads, which is a common method. The roads present many kinds of features in the remote sensing images such as gray features and spectrum features. This dissertation analyzed these road features and researched the methods of road extraction from remote sensing images. The dissertation detail contents as following:First, the development process of the multi-resolution analysis and the basic theory of the contourlet transform as well as the math expression of the remote sensing images were introduced. Also, the disadvantages of wavelet transform using in the image processing were analyzed. The spectrum features of main objects in the remote sensing images and some factors which influence the accuracy of the road extraction were analyzed.Second, a new algorithm for road extraction from remote sensing images was proposed based on nonsubsampled contourlet transform (NSCT). Firstly, the coefficients in different scales and different directions were obtained by decomposition using the NSCT. Then, calculated the maximum module within certain limit area of different scales and different directions, used the adaptive threshold to make the coefficients module binaryzation and deleted the small areas. Further, the points of the maximum module were extracted if they located in the extraction areas. Finally, the result could be achieved by snake tracing. Experimental results demonstrated that our algorithm outperformed other algorithms such as wavelet in accuracy and completeness.Third, using the advantages of the NSCT which are multi-scale and multi-directional as well as fully shift-invariant, an algorithm for road extraction from high resolution remote sensing images based on the NSCT and shape features was proposed. Firstly, the coefficients in different scales and different directions were obtained by decomposition using the NSCT, which were enhanced and reconstructed the image by the inverse transformation. Then the reconstructed image was segmented, and the linear and curve roads were obtained by using several object shape features. Further, the extracted roads were judged by spectral features. Finally, they were regulated by mathematical morphology. The experiments show the method of this study has a better effect on the high-resolution remote sensing images.