Integration of a pattern recognition method within raster geographic information systems.

摘要

This thesis presents a novel method for extracting user-defined geographic or spatial patterns from digital geographic images. In particular, this thesis develops and demonstrates a pattern recognition operation called Match integrated within Map•Factory™, a raster Geographic Information System software for the Macintosh Operating System. Match was designed, developed, programmed, and implemented by the author in a user-friendly framework for extracting spatially localized features from geographic images using pixels as the graphic elements. Through this technique, one can extract meaningful patterns formed by a composite of pixels. While there exists a resemblance between this method and the standard edge detection operations in their exploitation of two-dimensional images, Match is unique as it provides the user a more robust pattern extraction capability through specific search patterns as defined by the user. In Match, pre-defined patterns are designed by the user through a square pattern window having a minimum size of 3 x 3. This thesis further examines the performance of Match over standard edge detection operations and the cross-correlation method through quantified output comparison using applied examples. Finally, it is demonstrated through illustrated examples that by using a pattern recognition technique such as Match, raster Geographic Information Systems' pattern extraction capabilities are improved for various GIS spatial analytical tasks. As a pattern recognition operation, Match is still in early stages, but already it is possible to see the potential for use in many applications domains.