Automatic extraction of elevation information from the historical maps using machine learning techniques

Abstract

Historical maps are rich data sources of geographic data. However, without transcription, the map data is not machine-readable, meaning they cannot be used by modern GIS technologies. Currently, we still rely on tedious manual work to make the map machine-readable (such as crowdsourcing [34]). In this project, we present an automatic method to extract elevation data in the benchmarks from the Ordnance Survey 6-inch to one-mile historical map. Template matching and optical character recognition are used to locate and identify the numbers in the benchmark. The result shows that this method can efficiently extract the elevation data with minimum manual correction.