Next: Digitizing. Up: Geometric problems in Previous: Data Structure Tuning.

Generalization.

Since almost all GIS systems allow a user to zoom in on data, there is no intrinsic scale associated with the computer representation of a map. Nevertheless, there is an associated level of detail which determines a range of scales at which the data can be displayed without being too crowded or too sparse. Generalization is the process whose goal is to extend this range [23]. Several geometric problems fall under the heading of generalization:

LINE SIMPLIFICATION:
Approximate a polygonal line by one with less data (but hopefully the same information). This is has been the subject of much research in GIS, image processing, and computational geometry. Interesting problems remain for simplifying several lines (as in a contour map) or for demanding topological properties (e.g., forbidding self-intersection) from the result.
CLUSTERING:
Separate geometric objects can be aggregated and represented symbolically as a polygon or as a point.
DISTORTION:
Moving or changing the representation of geometric objects to enhance readability of the output. A railway that parallels a river may need to be displayed at lower resolution. A driver would prefer that a winding mountain road have some bends exaggerated on a map rather than being simplified to a line segment.

The more advanced generalization tasks are difficult to define a precise geometric optimization problems. Automating these tasks is often seen as an application for artificial intelligence techniques and ``knowledge-based'' heuristics. Computational geometry can offer not only efficient implementation of lower-level primitives, but also structures such as Voronoi diagrams and constrained triangulations that provide a more continuous model of space.


seth@graphics.lcs.mit.edu