Extracting Composite Cartographic Area Features in Low-Quality Maps

Abstract

We describe an extraction method for area features defined by composite cartographic elements and derived from historical, manually produced maps of low graphical quality. Composite elements appear in many topographic maps of the 19th and 20th century which provide unique information about the landscape of the past. We develop a method from prior research for extracting forest areas from the historic Siegfried Map, on which they are represented by sets of circular forest symbols within boundary regions. First, a prototype search identifies forest symbols characterized by a combination of geometric attributes of connected components, shape descriptions of the local neighborhood, and patterns formed by similar graphic elements. Next, the complete set of forest symbols is iteratively determined by testing forest symbol candidates in the vicinity of the prototypes if they belong to a group of sufficient other candidates. Finally, spatial expansion determines the forest net area described by the composition of recognized forest symbols and continues to fill gaps between forest areas and their boundaries, as well as larger objects within the forest area. Automated extraction from three map pages resulted in an accuracy of K = 95 percent, which indicates high robustness for automated processing of entire map series. The new approach presented in this paper represents a general methodological framework for the extraction of area features from composite map elements in low-quality maps.