Amalgamation in cartographic generalization using Kohonen's feature nets

Empirical observations of the way cartographers deal with generalization problems lead to the hypothesis that they first detect patterns of anomalies in the cartographic data set and then eliminate anomalies by transforming the data. Automatically identifying patterns of anomalies on the map is a difficult task when using GIS functions or traditional algorithmic approaches. Techniques based on the use of neural networks have been widely used in artificial intelligence in order to solve pattern-recognition problems. In this paper, we explore how Kohonen-type neural networks can be used to deal with map generalization applications in which the main problem is to identify high-density regions that include cartographic elements of the same type. We also propose an algorithm to replace cartographic elements located in a region by its surrounding polygon. The use of this type of neural network permitted us to generate different levels of grouping according to the chosen zoom-scale on the map. These levels correspond to a multiple representation of the generalized cartographic elements. As an illustration, we apply our approach to the automatic replacement of a group of houses represented as a set of very close points in the original data set, by a polygon representing the corresponding urban area in the generalized map.