CoChat

The mapKurator System: A Complete Pipeline for Extracting and Linking Text from Historical Maps

Open access
Kim, Jina, Li, Zekun, Lin, Yijun, Namgung, Min, Jang, Leeje, Chiang, Yao-Yi · 2023 · Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems

The DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index; none of the 15 sources its claims cite is checked for support yet.

Abstract

Scanned historical maps in libraries and archives are valuable repositories of geographic data that often do not exist elsewhere. Despite the potential of machine learning tools like the Google Vision APIs for automatically transcribing text from these maps into machine-readable formats, they do not work well with large-sized images (e.g., high-resolution scanned documents), cannot infer the relation between the recognized text and other datasets, and are challenging to integrate with post-processing tools. This paper introduces the mapKurator system, an end-to-end system integrating machine learning models with a comprehensive data processing pipeline. mapKurator empowers automated extraction, post-processing, and linkage of text labels from large numbers of large-dimension historical map scans. The output data, comprising bounding polygons and recognized text, is in the standard GeoJSON format, making it easily modifiable within Geographic Information Systems (GIS). The proposed system allows users to quickly generate valuable data from large numbers of historical maps for in-depth analysis of the map content and, in turn, encourages map findability, accessibility, interoperability, and reusability (FAIR principles). We deployed the mapKurator system and enabled the processing of over 60,000 maps and over 100 million text/place names in the David Rumsey Historical Map collection. We also demonstrated a seamless integration of mapKurator with a collaborative web platform to enable accessing automated approaches for extracting and linking text labels from historical map scans and collective work to improve the results.

Overview

CoChat Summary

TL;DR

mapKurator is an end-to-end system that extracts, post-processes, and links text labels from large historical map scans, outputting GIS-ready GeoJSON. It was deployed on over 60,000 David Rumsey maps and over 100 million text/place names, with integration into the Recogito collaborative annotation platform.

Key findings

  • The system provides six modules—image cropping, patch text spotting, patch-to-map merging, post-OCR, geocoordinate conversion, and entity linking—plus a semantic type recommendation API.
  • Deployment processed over 60,000 maps and over 100 million text/place names in the David Rumsey Historical Map Collection; it also processed thousands of Sanborn Fire Insurance maps and historical Ordnance Survey maps.
  • Output is GeoJSON containing bounding polygons, recognized text, post-OCR labels, and linked entity identifiers (e.g., OpenStreetMap IDs), enabling GIS editing and full-text search via Luna Imaging.
  • The semantic type recommendation API uses a fastText model pretrained on Wikipedia with 240 standardized Schema.org types; EntityLinker links labels to OpenStreetMap, Wikidata, or historical gazetteers.

Methods

  • Design: an end-to-end pipeline with modules M1–M6: crop high-resolution maps into tiles (e.g., 1000 × 1000 pixels), detect/recognize text with a Deformable DETR/TESTR-based text spotter trained on synthetic scene images, a synthetic map dataset, and real annotated images, merge patch predictions, run post-OCR via edit distance against an OpenStreetMap-derived vocabulary, convert image coordinates to geocoordinates using map metadata and GDAL, and link to external knowledge bases.
  • Data/sample: deployed at the University of Minnesota on the David Rumsey Historical Map Collection (>60,000 maps, >100 million text/place names), plus thousands of Sanborn Fire Insurance maps and Ordnance Survey maps.
  • Analysis: system description and demonstration; no quantitative accuracy evaluation or benchmark results are reported in the paper.

Limitations

  • Stated: automated results can still be imperfect; future work plans to add multiple-language processing, interactive machine learning from user annotations, and support for various ontologies and file formats.
  • Stated/evident: the current EntityLinker retrieves candidate geo-entities only from OpenStreetMap satisfying substring and geometry criteria, limiting linkage scope.
  • Evident: as a demo paper, it reports deployment scale but no precision, recall, or error rates for text spotting, post-OCR, or entity linking.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3589132.3625579

Cites

20 of 20 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 System workflow with mapKurator architecture; mapKurator consists of six modules (abbreviated as M#) and semantic type recommendation API. O# denotes output files and I# with dotted lines are the inputs for each module.
Figure 1 System workflow with mapKurator architecture; mapKurator consists of six modules (abbreviated as M#) and semantic type recommendation API. O# denotes output files and I# with dotted lines are the inputs for each module.CC-BY-NC-ND · From the PDF
Figure 2 A text label example of the mapKurator 's output file in GeoJSON format.
Figure 2 A text label example of the mapKurator 's output file in GeoJSON format.CC-BY-NC-ND · From the PDF
Figure 3 A list of semantic types retrieved from recommendation API while typing 'hote'.
Figure 3 A list of semantic types retrieved from recommendation API while typing 'hote'.CC-BY-NC-ND · From the PDF
Figure 4 Visualizations of mapKurator 's results from the David Rumsey Map Collection (web interface powered by Luna Imaging). The left figure presents the search results for 'Mississippi', and the right figure shows the search results for 'Hamburg'. Each tile represents a detected and recognized text label from the 60,000 processed georeferenced maps. Note that these text labels can be in varying orientations and font sizes. A toggled text label on the right figure shows the full map image with pinned location (i.e., orange arrow).
Figure 4 Visualizations of mapKurator 's results from the David Rumsey Map Collection (web interface powered by Luna Imaging). The left figure presents the search results for 'Mississippi', and the right figure shows the search results for 'Hamburg'. Each tile represents a detected and recognized text label from the 60,000 processed georeferenced maps. Note that these text labels can be in varying orientations and font sizes. A toggled text label on the right figure shows the full map image with pinned location (i.e., orange arrow).CC-BY-NC-ND · From the PDF

Verification

Claims the paper makes

0 Valid · 3 Not valid · 11 Not checked · 19 Own results and interpretations

4 more claims

Create a free account to see all 33 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 22
  • It may state two separate claims as one. 10

Create a free account to see these claims.

Citations

Cites

Show all 20 references

Cited by

Metadata

Authors
Kim, J., Li, Z., Lin, Y., Namgung, M., Jang, L., & Chiang, Y.-Y.
Affiliations
University of Minnesota (US)
Venue
Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems
Publisher
ACM
Published
2023-11-13
Pages
1-4
OpenAlex
W4390100393
Cited by
33
References
18
Citations by year
2026: 12 · 2025: 13 · 2024: 6 · 2023: 2
FWCI
1.89
Type
conference paper
Language
en
Open access
Locations
  • ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (SIGSPATIAL GIS) (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by-nc-nd)
  • arXiv (Cornell University) (cc-by-nc-nd)
Funders
  • Nvidia
  • University of Minnesota
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Handwritten Text Recognition Techniques
  • Social Sciences › Arts and Humanities › Literature and Literary Theory › Digital Humanities and Scholarship
  • Social Sciences › Geography, Planning and Development › Historical Geography and Cartography
Keywords
historical maps, text extraction, text linking, optical character recognition, place name recognition, GeoJSON, FAIR principles, geographic data

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-19, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-19 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.