Research On Theory, Methods And Applications Of Geometry Similarity Measurement For Spatial Data

The similarity is the basis of the human perception, identification, classification, reasoning and other cognitive activities. The similarity measure of spatial data is not only the key technologies of map navigation, spatial data matching, spatial data fusion, spatial data updating, spatial data retrieval and similarity query, spatial data clustering and anomaly detection, but also it has very important significance in exploring the variation laws of geographic entities in different period and implementing dynamic prediction. This dissertation focuses on how to establish and apply geometric similarity measurement model for multi–source, multi–scale and multi–temporal spatial data. The main contributions are as follows:1. The dissertation summarizes the research actuality of similarity science and engineering, and geometric similarity applications, actuality, challenges and solution in the field of computer vision and geography information science are also present. This dissertation thinks that the theory basis of spatial data geometric similarity measurement model establishment is what similarity science and engineering includes interrelated theory, and the key algorithms are based on image and spatial relations processing algorithms, but the particularity of spatial data must be take into account.2. The dissertation discusses the basic definition, property and classification about similarity based on psychology, cognition science, noetic science and system science, and then the similarity measurement distance–based, similarity measurement system–based and several typical geometric similarity measurement models are also introduced. Based on the above basic theory, this dissertation presents the scientific meaning, concept, basic property for spatial data similarity, and the causes of spatial data similarity are analyzed from the geography entity changes. Then the dissertation analyses and defines spatial data similarity from the system point of view, i.e. the spatial data was seen as a system, which is divided into many hierarchies. The logic relations of system–element-feature about are spatial data similarity explained. Based on system analysis, this dissertation conducts a hierarchical classification and formal definition mathematically, and the important conclusion that the similarity measurement methods for spatial data are the organic integration of similarity measurement distance–based and similarity measurement system–based is got.3. The dissertation establishes kinds of geometric similarity measurement models between the vector objects, and researches on the models applications in spatial objects matching, conflation and cartographic generalization, includes: this dissertation introduces the basic idea and method of point pattern matching and proposes similarity matching method of between spatial point group objects based on KL feature, geometric similarity measurement model of point group objects is established. Finally, the dissertation applies the optimum interpolation method to conflate the multi-source geography spatial data based on point group matching, compared with other methods, this method can improve the data precision; the dissertation presents a method of curve similarity measurement based on average Fréchet distance(the method can also identify neighborhood points between homonymy curves). This dissertation proposes an algorithm for feature matching from network data at different map scaled based on similarity measure. The whole strategy of matching is the first pre-matching of nodes and arcs, followed by accurate matching through similarity of node-arc topologies and discrete Fréchet distance. The matching process combines the matches in geometry, semantics, topology, nodes and arcs effectively. Based on matching, line objects are conflated; the dissertation establishes universal measure model of geometry similarity for multi-scale spatial data based on multilevel chord length functions and center distance functions. These functions can describe geometry shape from entirety to part gradually. This dissertation improves the traditional Hausdorff distance based on the statistic Gaussian mode. The enactment of every criteria threshold value in the measure model of geometry similarity is solved by introducing relevance feedback techniques. At last, the model is applied in data matching of different scales and similarity measure of spatial object simplification.4. The dissertation researches on how to apply active contour model and existing vector data extracting and updating area and linear objects based on the geometry similarity between the vector data contour and image data edge contour. The basic theory, solving methods, advantages and disadvantages of the traditional active contour model are introduced. This dissertation presents a new active contour model, which is used to extracting area water body and linear road from Remote Sensing Images. The new active contour model is added to the image gravitation potential energy based on object-background gray value and the similarity restriction potential energy based on discrete curvature, the purpose is to improve the model constringency speed and noise immunity, and to avoid the noise on the curve point of attraction and disturbance. The model use prior information of vector data adequately to compute correlative parameter adaptively. The balloon power is added into original line active contour model to extract and update linear objects. At the same time, The extraction precision evaluating model is established based on similarity measurement, and the process of solving the model is given based on greedy algorithm.5. The dissertation researches on the establishment of topological relations similarity measurement model, direction relations similarity measurement model and five–tuples similarity measurement model. Based on 9–intersection classification and description of topological relations, the distance values between various topological relations are computed and the simple topological relations similarity measurement model between objects set is established. This dissertation establish similarity measurement model of complex topological relations by using the strategy of decomposing– combination based on simple topological relations similarity measurement model; Goyal's conceptual neighborhood graph is improved and the normal direction relations similarity measurement model is established based on direction relations describing methods, such as 1:N, N:N and M:N direction relations similarity. Then direction relations similarity measurement model between objects set is established, and finally, this dissertation establishes five–tuples similarity measurement model based on quantity, dimension, geometry, topological relations and direction relations similarity measurement model. The reasoning process of geography event is presented based on mixed similarity measurement model and the database logical expression frame and the detection and storage flow for geographical event is designed. At last, the dissertation's research are used in map making and geography information updating system, and the basic content and main function of experiment system are also introduced.