Accurate segmentation of land regions in historical cadastral maps

Abstract

Historical cadastral maps are valuable sources for historians to study social and economic background of changes in land uses or ownerships. In order to conduct large-scale historical research, it is essential to digitize the cadastral maps. As being established in antiquity, however, they suffer from significant noise artifacts attributed to hand-drawn cartography. In this paper, we propose a novel method of extracting land regions automatically in historical cadastral maps. First, we remove grid reference lines based on the density of the black pixel with the help of the jittering. Then, we remove land owner labels by considering morphological and geometrical characteristics of thinned image. We subsequently reconstruct land boundaries. Finally, the land regions of a user’s interest are modeled by their polygonal approximations. Our segmentation results were compared with manually segmented results and showed that the proposed method extracted the land regions accurately for assisting cadastral mapping in historical research.