Active Learning for Classifying Template Matches in Historical Maps – Datasets


On this page, you find the data sets used in the paper Active Learning for Classifying Template Matches in Historical Maps. The data sets have been created by applying template matching on bitmap images of historical maps. The table below provides basic information about the individual data sets, including the respective template used in template matching.

Every dataset can be downloaded as a .csv file, each of which contains 1.000 samples. The samples consist of the following three attributes:

rank
the rank of the sample, ordered from 0 to 999 by score [integer]
score
the template matching score of the sample (ratio of pixels equal to template) [continuous]
correctness
whether the sample shows one of the accepted characters [boolean]

The attribute to be predicted is correctness. Note that for some templates, there are additional characters that were considered correct besides the character shown in the template. This is mostly because the character in the template is visually contained in the additional characters on the specific map. For details, see the discussion in the paper.

# Historical Map Library Identifier Template Accepted Characters
1 Carte Topographique D'Allemagne, Daniel Adam Hauer (1787) 36/A 1.16-41 b, h download
2 Franciae Orientalis, Sebastian von Rotenhan/Abraham Ortelius (1570) 36/A 20.39 a, g, d download
3 Franciae Orientalis, Sebastian von Rotenhan/Abraham Ortelius (1570) 36/A 20.39 e download
4 Circulus Franconicus, Frederik de Wit (1706) 36/A 1.17 a, g, d download
5 Das Franckenlandt, Sebastian von Rotenhan (1533) 36/G.f.m.9-14,136 a, g download
6 SRI Comitatus Henneberg, Johann Jakob Zinck and J. G. Küsel (1743) 36/A 1.13 n, m, h download
7 SRI Comitatus Henneberg, Johann Jakob Zinck and J. G. Küsel (1743) 36/A 1.13 e download
8 Circulus Franconicus, Frederik de Wit (1706) 36/A 1.17 download
9 Circulus Franconicus, Matthias Seutter (1731) 36/A 1.18 download

All data sets presented in the paper and on this page are based on historical maps from the Franconica collection of Würzburg University Library. We thank Hans-Günter Schmidt of Würzburg University Library for providing these scans maps and for providing practical use cases.


Contact the authors: Thomas van Dijk and Benedikt Budig, Chair of Computer Science I, Universität Würzburg, Germany. Impressum/Legal Notice