For several applications, it is interesting to have an idea of database contents. For instance in geo-marketing, a decision-maker could be interested in knowing the places where his products are not well sold. In other words, the summary of a database must include only the important items. The goal of this paper will be to present an international project aiming at generating visual summaries of geographic databases based on chorems. From years, chorems – as schematized representations of territories invented by a geographer named Brunet – are used to represent the more salient aspects of a territory. But in this methodology, the designer is supposed to know everything and to propose a schema. In contrast, in our project, the starting point is a geographic database representing information over a territory for which one wants to generate a drawing emphasizing only the salient aspects called patterns. So, two main problems exist, first to extract the salient patterns from spatial data mining, and the second visualizing the results after having done both a geometric generalization of the territory boundaries, and a semantic generalization of the contents. This paper will first present the problems of discovering patterns and selecting the more important ones (5  2). Then, we will explain how to make the geometric and semantic generalizations. After having presented a model of patterns, we will present the ChorML language used for both for storing and visualizing them.