Scans go through the same door
A page with no text layer is detected and rendered at the resolution of the embedded image, then read by a recognizer running in its own process on the same PC. The decision is made page by page, so a normal text PDF is never slowed down. Recognition covers Latin scripts (EN, DE, FR, ES, IT, PL, CS, SK, TR) and Cyrillic (RU, UA, BY). CJK, Arabic, Thai, Greek and Hebrew are not supported. Around 3 seconds per page is typical, so a 40-page statement takes roughly two minutes. Scanned batches often carry blank backs and separator sheets, so it pays to remove blank pages from PDF files before the run.