OCR

An old lease, a stack of receipts or a photographed page looks perfectly readable and yet Ctrl+F finds nothing in it. That is because the scan is a picture of writing: your computer sees pixels where you see words. Optical character recognition reads those pixels and turns them back into characters you can search, copy and paste.

This tool recognises English and Korean, and also offers Japanese and Simplified Chinese, each paired with English, though for those two the .txt is the result to rely on because the searchable PDF's text layer is set in a Korean and Latin font. Accuracy depends almost entirely on the source, so aim for about 300 dpi, pages that are straight rather than tilted, and even lighting without shadows across the gutter. Handwriting, stamps over text, heavily patterned backgrounds and faint photocopies all reduce the hit rate, and digits are the most expensive thing to get wrong, so check amounts, dates and reference numbers by eye.

You always get two files, the plain text and a searchable PDF in which each page is re-rendered as a picture at roughly 144 dpi with the recognised text in an invisible layer over it, so the document still looks like the scan but is not the original file. Recognition runs entirely in your browser, which means the engine downloads once on first use and processing time grows with the page count; a long document is better done in batches. If the PDF was created digitally rather than scanned, pdf-to-text will give you the exact characters with no recognition error at all.