Information Extraction based on the Concept of Geographic Context

State-of-the-art graphics recognition technologies for extracting geographic information from scanned map images are very labor intensive and do not scale well to process a large number of maps. Moreover, many historical scanned maps suffer from poor graphical quality due to bleaching of the original paper maps and archiving practices, and are ill-posed for traditional one-time training and recognition efforts. This paper introduces a novel context-based framework for automated recognition of geographic information from cartographic documents including historical scanned map documents of varying graphical quality. This recognition approach makes use of the fact that map content of the same area changes incrementally over time, and such dependencies can be used as geographic context to fully automate the recognition process. If successful, such a framework can be useful for extracting geographic information from whole map series at a national level and over long time periods enabling new forms of landscape research. We demonstrate this framework in a case study on exploiting contextual information about a map area to extract hotel symbols from a scanned paper map, automatically, and discuss the applicability of this framework using geographic layers as context.