Scanned PDF to EPUB: OCR, and what to expect from it

What OCR does and does not fix

OCR reads the pixels of a page and guesses the characters. It is good at clean printed text and bad at anything unusual: a broken letter, a stain on the page, a decorative initial, a footnote in small type. The output will contain typos, and no amount of downstream processing removes them completely. What OCR does fix is the thing that makes a scan unusable: the text becomes selectable and searchable.

Structure is the second problem

A book that was never digital has no chapter markers. The converter has to infer them. The usual signals are a line in a larger font than the body, a short line on its own, or a line that starts with a word like Chapter or Part. Where none of that is present you get flat text, and then the table of contents is empty, no matter how good the OCR is.

Practical order of work

1. Check the scan quality first. 300dpi, straight pages, no shadows.
2. OCR it, then read the first two pages of the result. The typos tell you how good the OCR was.
3. Look at the chapter list. If the structure is wrong, that is the file's fault, not the reader's.

Try it

Drop a scanned PDF below. The first 20 pages of a scan go through OCR free, with no account; the whole book is $5.