Deep Generation of Synthetic Training Data for the Automated Extraction of Semantic Knowledge from Historical Maps

As a prerequisite for backtracking and monitoring environmental changes, historical maps are invaluable as they precisely
capture the spatial reality of the past (Chiang et al., 2019). To avoid further deterioration of useful information that could
unlock the possibility for spatio-temporal analysis, often a digitization is performed, which results in a large number
of scanned historical documents stored in digital archives (Wu et al., 2023). However, in this case scanned historical
documents, e.g., maps or plans, are not semantically enriched which makes the underlying spatial information unreachable
(Chiang et al., 2019). In this context, it is desired to digitize heterogeneous historical map corpora in a way that their
semantic knowledge can be automatically extracted by machines, recently demonstrated through semantic segmentation
using different Deep Learning techniques (Arzoumanidis et al., 2023, Wu et al., 2023, Petitpierre et al., 2021). After
analyzing, vectorization allows to store extracted semantic knowledge in a machine readable format which sets the basis
for an effective and intuitive comparison with current map features, e.g., in a spatio-temporal analysis (Wu et al., 2023).