Optical Character Recognition

Optical character recognition (OCR) turns a picture of text into text a computer can search. The Library of Congress explains it for its Chronicling America newspaper collection: "OCR is an automated process that converts the visual image of text into machine-readable text. Computer software can then search the OCR-generated text for words, phrases, numbers, or other characters" (Library of Congress, Chronicling America research guide, "Improved Machine-Readable Text").

Why OCR makes errors

The Library states plainly that "OCR is not 100 percent accurate, and, particularly if the original item has extraneous markings on the page, unusual text styles, or very small fonts, the searchable text OCR generates will contain errors that cannot be corrected by automated means" (Chronicling America FAQ). Image quality matters too. Digitized microfilm "inherently includes a wide range of image quality," with quality derived "from the original newspaper, the original newspaper when it was microfilmed and associated deterioration, or the film itself."

Why it still helps

"Although errors in the process are unavoidable, OCR is still a powerful tool for making text-based items such as historical newspapers accessible to searching." Repetition rescues searches: "if OCR misreads one instance of a keyword in a passage, but correctly reads the second instance, the passage will still be found in a full-text search." A word that appears only once has no such backup.

OCR can be redone

"OCR technology has advanced significantly since the beginning of the NDNP," the Library notes. Its NDNP-Open-OCR project "uses Tesseract and custom post-processing steps" and "creates new ALTO XML and PDF files for NDNP batches." Reprocessing an archive this way is a Data Pipeline job.

Images without text

Not every digitized land record is text-searchable. BLM's General Land Office Records website "offers access to images of more than five million federal land title records issued since 1788," plus "images related to survey plats and field notes dating back to 1810." BLM's site instructions describe finding a document through a patent or Control Document Index search and then opening its image, and note that many entries "show dashes in the image column," meaning "These documents are not available here."

Reading handwriting and layouts with AI models is covered under Computer Vision and Deep Learning.

Why it matters for land and mining claim records

Recorded mining claim and land documents carry what the Library lists as OCR trouble: stamps and marginal notes (extraneous markings), old typefaces (unusual text styles), and small print. A full-text search that finds nothing in OCR text does not prove a record is absent. It may only mean the one instance of the name or number was misread.