Extraction Of Geographic Information Based On Maps And Remote Sensing Images

Geographic information is crucial for economy and social development. It is playing an important role in agriculture, forestry, water resources, land resources, geology and mineral resources, surveying and mapping, meteorology, ocean, urban planning, and defense. The development of geographic information database is time consuming, costly, complicated. Hence, it is imperative to acquire geographic information efficiently and intelligently. Some theory and techniques for extraction of geographic information from topographic maps and remote sensing imagery are studied in this dissertation. To be specific, we focus on the following aspects: map element segmentation, recognition and vectorization, road extraction in high resolution remote sensing imagery, vector data compression. The main research results obtained and novelty are summarized as follows.1. The importance of the research on geographic information extraction is illustrated. The essence and basic procedure of geographic information extraction are also analyzed. In addition, we review the history and state-of-the-art technologies in this field, pointing out some open problems still present.2. Considering shapes and scale features of map elements, some segment algorithms of map elements are proposed, which is based on multi-angled parallel computation theory of mathematic morphology. These algorithms combine the directional feature plane with non-isotropic operation and can be performed in parallel. With these algorithms, satisfactory segment results can be obtained.3. An approach to map numeric text recognition is proposed, which is based on radial basis function (RBF) neural network and recursive least squares training algorithm. The basic process includes: numeric text segmentation, feature vector extraction, neural network training, and pattern classification. It can be shown by an experiment that the proposed method has fast convergence speed while training, better robustness to noise and good classification ability.4. Different strategies of line tracing have been proposed for vectorization of map elements in binary images, including fading tracing, direction prediction tracing, parallel line cut tracing and contour tracing. These methods are particularly suitable for vectorization of lines and area border lines in maps. Since two map symbols, i.e., the gully and the cliff, have a great effect on the efficiency of automatic digitization of contour map plate, a strategy called revolving inner tracking is proposed for extracting gullies and cliffs, and corresponding extraction algorithms are presented as well. With this approach, gullies and cliffs can be extracted in a highly automatic manner.5. A new semi-automatic vectorization approach for color topographic maps is proposed based on sliding window image segment and sequential line tracking. A line feature is vectorized by adding a sliding window on it, self-adaptive image segment, thinning, and sequential tracking. The current line feature in the window is segmented by using color space conversion, K-means clustering and region growing. Sequential tracking is performed through tracking, fading, direction judgment and cross point processing. Experimental results show that compared with existing methods, better accuracy and robustness can be obtained, especially for color map images with lower contrast and lower SNR.6. Two semi-automatic approaches to extracting main road from high-resolution satellite images are presented, i.e., active window line segment matching for road centerline extraction and total rectangle matching for straight road extraction. The first scheme extracts road centerlines by using template window, threshold segmentation, line segment matching and an improved sequential similarity detection algorithm (SSDA). The second one convert straight road extraction to extraction of rectangle with specific direction and width. A road is matched from the inside to the outside to meet the optimization criterion by changing the threshold of image segmentation, the width and direction of the rectangle. Experimental results demonstrate that these two methods can extract main roads effectively with good robustness to noise.7. A new model based on total least squares for vector data compression is presented. Algorithms for polyline and polygon data compression are proposed. With the polyline data compression method, more precise vector data can be obtained in comparison with the popular Ramer-Douglas-Peucker (RDP) algorithm. The compression ratio obtained is very close to that of the RDP algorithm. The compressed data can represent the original data with improved precision. With the polygon data compression method, high precision data without redundant points can be obtained.8. An efficient and practical data acquisition system for geographic information has been established by using the proposed extraction models and methods. The acquisition of geographic information can be performed accurately, quickly, and conveniently. The system is being deployed in the production of 1:50000 geographic information data. Satisfactory results have been obtained.