Automatic interpretation of scanned maps
Keywords:
vectorization, GIS, artificial neural network, graphics recognition, pattern recognitionAbstract
To create a spatial database for some GIS applications, it is a big challenge to recognize all the simple and complex map objects automatically on scanned maps. This research field is generally referred as map interpretation. The first part of the present study shows the place of map interpretation concerning related topics of document analysis, and discusses interpretation systems representing main approaches in the field. The second part gives a detailed discussion of our own interpretation system called MAPINT which has special support from Hungarian cadastral maps. Processing starts with an affine coordinate transform followed by raw vectorization, the result of which is converted into a vectorgraph format. All recognition steps are performed on this format. Recognized objects are: dashed lines, house numbers and parcel numbers, connection signs (a special notation of Hungarian maps expressing the relationship between a building and a parcel), null-circles (denoting measured points), building and parcel polygons. The study is closed with a comparison and evaluation of different systems.