Quick answer
A scan is an image. There is no text in it to translate, so OCR must read the page first — which means recognition accuracy, not the translation model, sets the ceiling on quality.
How translating a scanned document works
- OCR converts the page image into machine-readable text
- The recognised text is checked, because every OCR error is about to be translated as if it were correct
- Translation runs on the text layer
- The layout is rebuilt so the translation sits where the original text was
What OCR gets wrong, and what it costs you
- Recognition errors are amplified, not caught: a misread word becomes a confidently mistranslated word
- Tables, formulas, and symbols are the first things lost — they often survive as plain text with the structure stripped
- Handwriting, stamps, and low-contrast scans drop out entirely
- Photographs taken at an angle or in uneven light lose whole lines
- If the scan already carries a hidden OCR layer, you can end up with two layers of text
How to get a better result from a scan
- Scan flat and straight at 300 DPI or higher, rather than photographing the page
- Flatten a multi-layer scan with Flatten PDF before submitting it
- Spot-check the numbers first — figures, dates, and part numbers are where OCR errors do the most damage
- Use the original digital file instead whenever one exists; it will beat any scan
How quota works for scanned pages
Scanned pages draw on the free OCR allowance in addition to your page quota, so a scanned file consumes more than a text file of the same length.
Related tools
Bottom line
Translating a scan is really two jobs: reading it, then translating it. Improve the scan and the translation improves with it.