The Deed Machine is a multi-language set of tools that use OCR and crowdsourced transcription to identify racially restrictive covenant language, then map the results.
Racial covenants are clauses that were inserted into property deeds to prevent people who are not white from buying or occupying land. As of June 2024, Mapping Prejudice volunteers have helped to map more than 52,000 covenanted properties across 3 states, with more on the way soon.
The Deed Machine was created at Mapping Prejudice at the University of Minnesota Libraries. Current collaborators include Michael Corey, Suleman Diwan, Justin Schell, and the University of Minnesota Libraries IT staff.
This is the code repository for the Django component of the Deed Machine, racial_covenants_processor. See the full docs for more information about other components.
- geos and gdal
- proj
- pandas and geopandas
- pipenv
- PostGIS/PostgreSQL
- AWS SAM for lambdas (separate repos)
- libmagic (mostly to silence panoptes/zooniverse warnings)
Often deed images are stored on a local machine or network drive, and it's not feasible or efficient to move them. This standalone uploader is designed to avoid the user having to do a full install on this computer.
The individual lambda functions that make up the OCR, term search and web image optimization processes are in separate repositories:
- mp-covenants-split-pages
- mp-covenants-ocr-page
- mp-covenants-term-search-basic
- mp-covenants-resize-image
- mp-covenants-fake-ocr
python manage.py shell_plus --notebook