Skip to content

introduction.md

Latest commit

 

History

History
59 lines (41 loc) · 3.54 KB

introduction.md

File metadata and controls

59 lines (41 loc) · 3.54 KB

mapKurator is a fully automatic pipeline developed by the Knowledge Computing Lab at the University of Minnesota to process a large number of scanned historical map images. Outputs include the recognized text labels, label bounding polygons, labels after post-OCR correction, and a geo-entity identifier from OpenStreetMap.

Model Summary

  • Green boxes: Modules in the pipeline
  • Blue boxes: Inputs of the modules
  • Orange boxes: Outputs of the modules

image

Model Details

  • ImageCropping divides large map images (>10K pixels) into map image patches (1000 x 1000 pixels) for PatchTextSpotter to process.

  • PatchTextSpotter automatically detects and recognizes the text labels on map image patches. The mapKurator system offers two state-of-the-art approaches for text spotting, where the models are trained with synthetic datasets and human annotations.

  • PatchtoMapMerging merges the patch-level spotting results into map-level. The output polygons can be loaded in QGIS for visualization, and they should be aligned with the ungeoreferenced map image. Note: JP2 files should be converted to JPEG before loading in QGIS.

  • PostOCR aims to verify the output and correct misspelled words from PatchTextSpotter using a dictionary created from OpenStreetMap. PostOCR identifies a word's candidates using the fuzzy query function from an elasticsearch index of the OpenStreetMap dictionary, which contains the place name attribute from OpenStreetMap. Then PostOCR picks one candidate based on the word's popularity in the OpenStreetMap dictionary.

  • GeocoordinateConverter converts the bounding polygons of text labels from the image coordinate system to the geocoordinate system. Note: The polygons in both coordinate systems are saved in the output.

  • EntityLinker links each map text to all the candidate geo-entities in OpenStreetMap. The entity linking retrieves candidates that satisfy both criteria: 1) the suggested word (i.e. output from PostOCR) is a substring of the candidate geo-entity's name 2) the geocoordinates of a geo-entity is within the map boundary. (Geo-coordinates are obtained from GeocoordConverter)

Cite

@inproceedings{li2020automatic,
  title={An automatic approach for generating rich, linked geo-metadata from historical map images},
  author={Li, Zekun and Chiang, Yao-Yi and Tavakkol, Sasan and Shbita, Basel and Uhl, Johannes H and Leyk, Stefan and Knoblock, Craig A},
  booktitle={Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  pages={3290--3298},
  year={2020}
}

@inproceedings{10.1145/3589132.3625579,
    author = {Kim, Jina and Li, Zekun and Lin, Yijun and Namgung, Min and Jang, Leeje and Chiang, Yao-Yi},
    title = {The MapKurator System: A Complete Pipeline for Extracting and Linking Text from Historical Maps},
    year = {2023},
    isbn = {9798400701689},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3589132.3625579},
    doi = {10.1145/3589132.3625579},
    booktitle = {Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems},
    articleno = {35},
    numpages = {4},
    keywords = {linked data, automatic system, text spotter, historical maps},
    location = {, Hamburg, Germany, },
    series = {SIGSPATIAL '23}
}

License

CC BY-NC 2.0