OCR tips: how to get accurate text from scanned documents
Why OCR makes mistakes and how to prevent them: scan resolution, lighting, rotation, contrast and file preparation that dramatically improve text recognition.
By Fileixa editorial team, Creates Online LLC · Published · 5 min read
What OCR sees
Optical character recognition looks at the shapes in an image and decides which letters they are. It does not understand the page the way you do. Anything that makes letter shapes harder to separate from the background, such as blur, shadows, low resolution, a tilted page or a patterned background, turns into wrong characters in the output.
Scan settings that matter most
In order of impact:
- Resolution: 300 dpi is the sweet spot for normal-size text. 150 dpi loses small print; above 400 dpi rarely helps and makes files huge.
- Straightness: rotate pages upright and keep them flat. OCR engines handle small tilts, but sideways or upside-down pages fail badly.
- Contrast: black text on a white background works best. Greyscale scans are fine; very dark or yellowed pages benefit from higher brightness.
- Focus and lighting for phone photos: shoot in even daylight, avoid your own shadow, hold the phone parallel to the page.
Documents OCR struggles with
Handwriting, stylised or decorative fonts, text over photos, very small print, and complex tables are the usual problem areas. For tables, it is often faster to OCR the text and rebuild the table than to fix a badly recognised one. Always proofread numbers such as amounts, dates and account numbers, where a single wrong digit matters most.
Preparing the file
Rotate any sideways pages first, remove blank pages, and crop away dark borders from photocopies. If the PDF is a mix of real text and scanned pages, you only need to OCR the scanned part; text that is already digital is extracted perfectly without OCR.