Research On Road Extraction Method In High Resolution Remote Sensing Images Based On The Snakes Model

Road network is very important and basic geographic information. In recent years, with the development of high resolution satellite remote sensing images (such as IKONOS and QUICKBIRD, ORBVIEW etc.), high resolution images from extracting the space information and attribute information applying the high resolution images has become a reality. But how to extract the road network automatic from the high resolution remote sensing images is a very challenging task. Therefore, it has become one of the hot problems in some fields, such as remote sensing, computer vision and image understanding.Basing on the active contour model (Snakes) theory, the paper researches the road extraction of the high resolution remote sensing images. According to the characteristics of roads in the high resolution remote sensing images, by improving the existing method, the paper draws an active contour model that is more suited to the road extraction, and completes the road extraction further. The main elements are as follows:(1) Basing on the deeper analysis of the existing semi-automatic Snakes road extraction methods, the paper draws the basic ideas of the road extraction that also uses the Snakes: the stage of obtaining the initial conditions of the road and the stage of using the initial conditions to track and process the road network.(2) The initialization contour of traditional Snakes used the individual pixels (control points) for its basic elements, which makes the information that is provided to the high level to control and option too numerous and disorganized. In the stage of the road initialization, based on the linear detection and searching for the parallel lines of the road ,the paper organizes the road information initially, as a result of the Snakes'basic elements and thus designing energy function, so that the model could be enhanced on the stability and efficiency.(3) It is efficient to detect line using the straight linear structural feature in the digital images, while certainly fracture resistance. In the detected lines, the parallel lines could be extracted by comparing the gray uniformity between the adjacent lines, which provides the initial condition to the road extraction for the active contour model.(4) In the road network tracking and processing stage of using the initial condition, the paper focuses on solving the Snakes energy function designing. The elastic energy in the traditional energy item is improved, and view the road edge feature of the high resolution remote sensing images, the new parallel restriction is increased.(5) The paper is improved the method of identifying the iterative region. With the introduction of Snakes growing thinking, we implement a strategy that extending and optimization are completed on the same time to enable it to extract the complete road edges line better.Based on the experiments of 1-meter resolution IKONOS image, the results show that this method can be used to extract the road edge information of the high resolution images.