Automatic content‐based georeferencing of historical topographic maps

Abstract Old maps are more than a cultural artefact: they are data. Data about the past still hold value for science and decision‐making today. Libraries and archives have come a long way in digitising their inventories of thousands, sometimes millions, of historical maps using high‐resolution scanning. Unfortunately, even with digital images the rich spatial and semantic information is inaccessible for people without a strong background in history and cartography. Only with georeferencing can historical maps be used in GIS and thus processed and compared with modern geospatial data. We introduce content‐based image retrieval to automatically localise and georeference map images from topographic map series. We align the maps by extracting a subset of their symbols and cross‐referencing them with online reference data from OpenStreetMap. We demonstrate our method with the Karte des Deutschen Reiches at 1: 100,000 scale, obtaining 96% correct location predictions and a median georeferencing error of 101 m.