Recognition of Cartographic

Sushil Bhattacharjee Gladys Monagan Institut f ur Informationssysteme Swiss Federal Institute of Technology (ETH) ETH-Zentrum, CH-8092 Zurich, Switzerland ABSTRACT A hybrid (statistical/structural) approach is presented, for scaleand orientation-invariant recognition of multi-component cartographic symbols. A decision-tree classi er (DTC) is used to identify the shapes of the individual components of a symbol. Structural matching is then used to determine the type of symbol under consideration. INTRODUCTION Machine-interpretation of cartographic maps has come to occupy an important place in the burgeoning document-image-processing industry. Several comprehensive collections of papers on this topic are now available [1]. Cartographers often use prede ned symbols to convey such `meanings' associated with logical structures represented in maps. Recognition of cartographic symbols is, therefore, an important aspect of any map-interpretation system. In this paper, we describe a method for recognizing cartographic symbols that has been developed for processing digital images of land-registry maps in Switzerland. These maps are basically line drawings which identify the various regions of an urban neighborhood. The proposed method for symbol recognition operates on bilevel images. It is independent of the size and orientation of the hand-drawn symbols, and is also independent of the scanning resolution of the input image. Figure 1 presents an overview of the symbol-recognition approach proposed here. The di erent stages of the owchart shown Scanned Image