Geographia Technica, Issue no. 1/2013, pp. 34-46

A comparative study of geometric transformation models for the historical ‘Map of France’ registration

Pierre-Alexis HERRAULT, David SHEEREN, Mathieu FAUVEL, Claude MONTEIL, Martin PAEGELOW

ABSTRACT: It is widely recognized that present-day biological diversity may reflect past land use and landscape features. Understanding the historical background is essential to explain the ecosystem functioning of landscapes today. Numerous historical spatial data have been used to reconstruct these pathways but resulted in problems of sources, formats and supports. One major problem is how to incorporate historical data in current coordinate systems to enable them to be compared with each other and with current sources. Previous works on the registration of historical maps already demonstrated the superiority of local methods (e.g. a Delaunay-based method) over global methods (like polynomial mapping models) because of the existence of local geometric distortions. The same studies also highlighted the importance of selecting ground control points with a homogeneous spatial distribution to improve the accuracy of registration. Furthermore, while kernel-based methods have already proved their efficiency for many other applications, they have rarely been used for map registration even though they provide an interesting alternative to conventional methods. In this paper, we present a comparative study of various geometric transformation methods to register an excerpt of the historical ‘Map of France’ (an Ordnance Survey map) dating from the 19th century. We compare the performance of several global and local methods with kernel-based methods (Gaussian and polynomials) and analyze the impact of the number of control points, their nature and their spatial distribution on the quality of registration. A protocol was developed to apply the transformation models in various situations and to identify the best strategy for the georeferencing of historical maps.

Keywords: Historical map registration, geometric correction, kernel regression. .

Full article here