Based On High Resolution Remote Sensing Images Of The Active Contour Model For Road Extraction

With advantages such as a short cycle and an inexpensive way to obtain information, and rich information, satellite remote sensing images can quickly acquire information on the Earth's surface. In recent years, IKONOS, Quick Bird, Spot-5 and other high-resolution remote sensing satellites launched in succession, provided us with more and more content-rich satellite remote sensing data. However, the traditional manual interpretation and identification methods can't guarantee the efficiency in access to information and the discrimination accuracy, which largely limits the use of these resources. Therefore, it becomes an important developmental direction of remote sensing information processing to identify and understand objects in remote sensing images quickly and accurately,using image analysis technology.Roads have significant geographical and geometrical features, which are not only important basis geographic information, but also can serve as clues and reference to extracting other objects on the Earth's surface. Therefore, the road extraction has been a key research topic of the application of remote sensing images. According to the characteristics of roads in high-resolution remote sensing images, the thesis,based on full understanding and learning the existing road extraction approaches, proposes new road extraction methods by combining region growing with active contour model theory, and utilizes them for extracting roads.In this dissertation, the main research contributions are as follows:(1) Based on analyzing the existing initialization methods of the active contour model, the thesis proposes two new initial road location extraction algorithms:an adaptive region growing algorithm and a color region growing algorithm due to principal component analysis (PCA).The former can estimate the parameters of homogeneity criterion automatically in the growth process, according to the gray characteristic distribution of the object to be segmented. And the preliminary experiment shows that compared with the traditional threshold region growing algorithm, the method has a better ability to learn and adapt, and needs less manual intervention(in general, one seed point is enough). The latter, to make the most of the image color information while reducing the computational time, applies the PCA algorithm to extract the color feature of objects, and uses interval estimation to construct the corresponding homogeneity criterion in addition. The preliminary experimental results show that in the case of less human intervention, the algorithm can make the best of the color characteristics and complete the pre-segmentation of objects effectively, with the high time efficiency.(2) Combining the adaptive region growing algorithm with the GVF(Gradient Vector Flow)-Snake model, a new semi-automatic road extraction method is proposed in this dissertation and is employed to extract roads from high-resolution remote sensing images. In the method, the adaptive region growing algorithm is first applied to the preliminary road segmentation, and then mathematical morphology is utilized to eliminate disturbances inside and the ordered outline of the road in the grown image is obtained by a contour tracking operator. Finally, we use the outline as the initial contour of the GVF-Snake Model, and apply the model to tracking the road, achieving the final result of the road extraction. Experimental results show that with less manual intervention, the method can extract more accurate and complete roads from remote sensing images, and has a certain practicability and robustness.(3) A road extraction approach is presented in this thesis, which employs a color region growing algorithm together with the level set method to extract roads from remote sensing images. Firstly, the color region growing algorithm based on PCA is utilized for the preliminary road segmentation, and the mathematical morphology is applied to eliminate disturbances inside in the grown image. And then we use the roughly obtained road region to construct the initial level set function, which then evolves stably according to a level set evolution equation without re-initialization. Finally, a regularizing local edge algorithm based on level set without re-initialization is proposed, and utilized to regularize the local erroneous road curve due to obstacles, completing the final work of road extraction. Experimental results show that the method is efficient and practical for extracting complete roads from high-resolution remote sensing images, with less manual intervention and stronger anti-noise ability.(4) An improved variational level set method is suggested and applied to road extraction of high-resolution remote sensing images. The new model is a variational level set method which is adapted to extract objects of interest from complex backgrounds and is achieved by introducing such three terms into GACV (Geodesic-Aided CV) model as the object identification function constructed based on the color region growing algorithm, the color gradient flow computed according to the Beltrami framework, and a penalizing term which serves as a metric to characterize how close the level set function is to a signed distance function. Experimental results show that the model can effectively extract roads from high-resolution remote sensing images, considerably reducing the interference from non-road objects, and has a certain practicality.