Reference Glossary
OCR (optical character recognition)
Technology that extracts machine-readable text from an image or scanned document — a sign in a photo, text on a scanned page, a caption baked into a graphic — converting pixels into searchable characters.
Why it matters in a DAM
OCR is what makes a scanned print-collateral archive or a photographed product label searchable by the words visible in it, closing a gap that manual metadata tagging realistically never covers for a large legacy archive. Accuracy is not uniform across content, though: Tesseract, one of the most widely used open-source OCR engines, reaches roughly 98–99% accuracy on clean printed English text, but one benchmark measured its accuracy on scanned Turkish-language documents at only around 86%, and it degrades further on handwriting and low-resolution scans.
A worked example
Common mistake
Teams run OCR once at ingestion and never re-run it after an asset is cropped, rotated, or replaced with a corrected version, so the indexed text silently goes stale relative to what's actually visible in the current file.
OCR converts the visible text inside an image or scanned page into an actual character string a search engine can index — the difference between a scanned contract being a picture of words and being a document a keyword search can actually find. In a DAM, it’s typically run as a processing step at or after ingestion, with the extracted text stored as its own searchable field.
The value is largest for legacy or print-heavy archives: scanned brochures, photographed signage, old contracts, product packaging shots. None of that content would otherwise be searchable by its actual text without someone manually transcribing it, which isn’t realistic at archive scale.
Accuracy claims for OCR need to be read against the specific content, not treated as a flat number. Tesseract, a widely deployed open-source engine, is commonly cited at roughly 98–99% accuracy on clean printed English documents, but a benchmark study measured its accuracy on a set of scanned Turkish-language documents at closer to 86%, and separate research has shown accuracy dropping much further on handwriting and low-resolution or noisy scans. For anything legally or operationally significant — a contract clause, a compliance label — that variance is a reason to keep a human verification step rather than trusting OCR output outright.
Frequently asked
What does OCR actually convert, and why does that matter for search?
It converts the visible text inside an image or scanned page into an actual character string a search engine can index — the difference between a scanned contract being a picture of words and being a document a keyword search can actually find.
How accurate is Tesseract, a common OCR engine, on clean printed English text versus other content?
Roughly 98-99% accuracy on clean printed English documents, but one benchmark measured its accuracy on scanned Turkish-language documents at only around 86%, and accuracy drops further on handwriting and low-resolution scans.
Why is OCR especially valuable for legacy or print-heavy archives?
Scanned brochures, photographed signage, old contracts, and product packaging shots wouldn't otherwise be searchable by their actual text without someone manually transcribing them, which isn't realistic at archive scale.
What happens if OCR isn't re-run after an asset is cropped or replaced?
The indexed text silently goes stale relative to what's actually visible in the current file, since OCR is typically run once at ingestion and won't automatically re-process a corrected version.
Should OCR output be trusted outright for legally significant content?
No — for content like a contract clause or a compliance label, the accuracy variance across content types, down to roughly 86% or lower on some non-English or low-quality scans, is a reason to keep a human verification step.
Where in a DAM pipeline does OCR typically run?
As a processing step at or after ingestion, with the extracted text stored as its own separate searchable field that full-text search then queries against.
Sources
- OCR accuracy of Tesseract on clean English document images is commonly reported at around 98–99%, but on a scanned Turkish document dataset it was measured at approximately 86.41%. checked 2026-08-07 — Performance comparison study (BPAS Journals)