Investigative Tools

OCR Tools

Skip to main content
Table of Contents
< All Topics
Print

OCR Tools

Last updated: October 21, 2026

FOIA releases, court exhibits, old financial disclosures and agency records often arrive as scanned images. Optical character recognition (OCR) turns them into text you can search and quote. OCR output is a machine’s reading of a picture, so every number and name you publish from it needs checking against the image.

Tool License / access Best for
Tesseract Open source Bulk OCR; language packs; local control
Google Cloud Vision Commercial API Hard scans, mixed layouts
Amazon Textract Commercial API Forms and tables extraction
DocumentCloud (add-ons) Platform Hosted investigative docs with OCR pipeline

What each one does, per its documentation

  • Tesseract is an open-source OCR engine under the Apache 2.0 license. Since version 4 it includes a neural-network (LSTM) engine focused on recognizing whole lines, and it still supports the older character-pattern engine. It reads more than 100 languages out of the box. It runs locally, so sensitive documents never leave your machine.
  • Google Cloud Vision offers two OCR features. TEXT_DETECTION extracts text from any image. DOCUMENT_TEXT_DETECTION is optimized for dense text and documents, and it is the one to use for handwriting and for PDF or TIFF files.
  • Amazon Textract detects typed and handwritten text and extracts forms and tables as structured data. It also has dedicated APIs for invoices and receipts and for US identity documents. Its table extraction suits financial disclosures and contract schedules.
  • Document platforms do OCR as part of ingestion. ICIJ’s Datashare runs OCR on scans and images so they become searchable, and Google’s Pinpoint can turn scanned tables into spreadsheets. See Search and Document Platforms.

Uploading a document to a cloud API sends it to that provider. For leaked or source-protected material, use a local tool.

Quality checklist

  • Sample 10 random pages per scanner batch; measure character error rate subjectively.
  • Watch for column order swaps in newspapers and financial tables.
  • Store OCR version and language model in the research log.
  • Re-key every dollar figure, date and docket number you plan to publish by reading it off the image, not the OCR text.

Provenance

OCR output is machine-derived. Label uncertain readings when quoting in published analysis; prefer image snippet + transcript side-by-side for high-stakes claims.

Sources

Expanded with AI research and drafting assistance from the primary sources cited; editorial review pending.

Was this article helpful?
0 out of 5 stars
5 Stars 0%
4 Stars 0%
3 Stars 0%
2 Stars 0%
1 Stars 0%
5
Please Share Your Feedback
How Can We Improve This Article?