A practical guide to image and PDF OCR
Start with a readable scan
Use a clear, straight photograph with even light. Crop the page border and avoid shadows across letters. For printed books, a 300 DPI scan is a useful starting point. Enlarging an already blurred image cannot restore missing detail.
Choose the language of the text
English uses a dedicated recognizer. Hindi and Sanskrit use a Devanagari recognizer. Choose the language actually printed on the page. Sanskrit compounds, accents and uncommon conjuncts need careful proofreading; recognition confidence is not a promise of correctness.
Read columns in sequence
Auto layout looks for a central gutter. For a Sanskrit–Hindi book with two parallel columns, choose Two columns and set the divider near the visible gutter. Output reads the left column from top to bottom, then the right column. This is not a translation alignment tool.
Extract just the pages you need
PDF ranges use physical page positions, starting at 1, rather than printed book numbers. From 4 to 8 includes five pages. Blank To page means the last page. You may process up to 20 selected pages from a PDF of up to 1,000 pages.
Review and save before expiry
Select a result page, correct the text, then copy or download page.N.txt. The ZIP contains the original OCR for every selected page, using the original PDF page positions. Browser edits are included only in the individual text download. Sessions expire 10 minutes after upload begins, even while OCR is running.