Automated and semi-automated map georeferencing

ABSTRACT Historical maps contain a wealth of information not generally available, but they must be referenced to well-known coordinate systems for maximum use in spatial analysis. Existing georeferencing tools are essentially manual, requiring considerable data entry, much panning and zooming, and precise on-screen digitizing. Here we present alternative approaches based on pattern-matching and spatial computing intended to overcome the inefficiencies of standard tools. We also describe and make available two computer programs implementing the methods discussed. The first, designed for large-scale quadrangles, locates map boundaries, finds ground control points, and produces georeferenced images without operator assistance. Experiments show that quadrangle georeferencing can be reliably automated (88% success rate in our tests). A second program, developed for general maps at any scale, uses self-learning and other approaches to overcome most of the manual aspects of georeferencing. Both programs find control points with single-pixel accuracy, yield transform errors on the order of map linewidth, and can produce warped or unwarped images as desired.