Text in Scans and Pictures (OCR)

Text in Scans and Pictures

Opensolr reads the text printed in scanned PDF pages and in the pictures inside your PDFs. A scanned contract, a brochure made of images or a photographed label inside a PDF is found by its words, like any other text.

Every page of a PDF is its own search result, so a visitor lands on the exact page that answers the search.

How the pictures are read

• Each picture is read in the language of its page. When the page has no text to tell, in the language of the document; when that is unknown too, in English first, then again in the language the text turns out to be.
• Very small pictures, like logos, icons and bullets, are skipped.
• Only words read with confidence are kept, so stains and noise do not turn into words.
• A word read with one wrong letter can still be found: the read text also matches by sound.
• A picture with no text in it can get a short description of what it shows: Pictures described by what they show.

Where it happens

• Crawling: automatic, for every PDF Opensolr finds on your site, on every index. See Documents on your site.
• Data Ingestion: send a PDF by its URL with rtf:true, or upload the file. See Files and documents and PDFs page by page.
• Your own code: read the text in pictures and documents by API: Pictures and documents by API.

What it uses from your plan

• On a crawl, reading pictures is free.
• On Data Ingestion and by API, each picture read uses part of one AI request of the account that owns the index.
• A picture read before is free, and the plain text of a document never counts.
• When the monthly AI allowance runs out, pictures are still read; only the descriptions of pictures without text wait.

More: What site search uses from your plan.

See it live

Search inside scanned public-domain PDFs on the OCR search demo.