Research on Fundamental Theory and Technical Approaches of Automated Map Generalization

This paper concerns only the generalization of geoinformation which is stored in cartographic database or spatial database of GIS.The author's basic opinion is that in cartography and GIS there is a lack of essential and operational theory which could support the automatic generalization in digital environment by computer.The author proposed and implemented the following theoretical approaches to the automatic generalization: Generalization concerns two problems: scientific and technologic.The former investigates "what is the essence of", "why the necessity and possibility of" and "what is the object of"; the latter determines "when", "where" and "how" to carry out corresponding operations.Without true theory there is no possibility to develop correct technical methodology and to make it have universal meaning. "What is the essence of generalization" should be examined from two aspects: philosophic and technologic.From the first aspect to view, generalization is not owned by cartography and GIS themselves, but it is a universal methodology of human cognition.Therefore, from the macroscopic point of view, the carto/geo-genaralization should be considered as a particular situation of application of scientific cognition rules "abstraction/summarization (the synonym of term generalization)".From the second aspect to view, the author proposed the insist of geo-info transformation as the essence of generalization. What is the object of generalization? The author proposed the DLM view of generalization, i.e.the generalization object is none other than the geo-content of database.The DLM view of generalization is equivalent to model generalization. "When" can be interpreted as object own conditions to be evaluated from the viewpoint of generalization.It can be considered as microconditions. "Where" can be interpreted as object "context" relations to be evaluated from the viewpoint of generalization.It can be considered as mesosituations. "How" involves very abundant connotation: recognition of local proximity relation, recognition of global macrostructure and determination of when and where a given method of generalization should be carried out. The (5W+1H) scheme builds up the basic concept/theoretical model. For implementing the model mentioned above, the author proposed the following structurized measures: 1) Automatic segmentation of nature line features (e.g.rivers, shore lines) into homogeneous segments using qusi optimized statitstic division method.The line homogeneity can ensure the effectness of line generalization algorithms. 2) Creating the embedded convex hulls of scattered point objects.These hulls represent both global distribution characteristics and the inner structure ones of point object set.Therefore, they can serve as an optimal structure to support implementing an adaptive generalization.Here the Voronoi graph can be used to further differentiate the importance for fine evaluation of point objects. 3) Creating the tree structure for river and valley networks to determine the position of objects in the structure.It is able to give global evaluation of object.If a Voronoi graph of line objects is added, then a further differentiation of importance of line objects can be made up.The latter step can enforce the structureness of generalization. 4) Creating the tree structure of contour lines to represent the mutual relationships (inclusion relation——vertical adjacency, apposition relation——horizontal adjacency) which provide a powerful means to automatically obtain so called HQDEM(high quality DEM).Only with such HQDEM can true break lines (relief structure lines) be derived geographically.The latter is a key auxiliary information for structure generalization of landforms. 5) Establishing the general structurized generalization model. On the basis of above mentioned the author proposed a general structurized model of generalization.It consists of three submodels:Global selection model determines how many objects should be selected.This is a map/database "conception" model;Structured select ion model determines w hat/which objects can be selected.This is a map/database "configeration" model;Geometry processing model, here an entity oriented data (semantic, metric and relational information)handling is carried out.This is an object "composition" model. Using the f ractal geometry a very essential parameter——fractal dimension value of an object itself or of object set can be obtained.This parameter describes variation rate with the change of scale of observations in preserving self-similarity.Because fractal dimension value describes structure aspects both for individual object and object set, therefore it can be used in structurized generalization. The author proposed "regeneralization" which divides the recent generalization operato rs into two categories:information transformation (DLM) class and graphics representation (DCM) class.