How to Clean Up a Scan Before OCR (For Better Accuracy)

People blame the OCR engine when recognition comes out full of errors — rn read as m, 0 as O, whole words mangled into nonsense. Usually the engine isn't the problem. The image is.

OCR works by matching shapes against known letterforms. Give it a clean, high-contrast page and it's remarkably accurate. Give it a grey, shadowed, soft-focus phone photo and it's guessing. Ten seconds of preparation before you run OCR can be the difference between a searchable document and a text layer full of garbage.

This guide covers preparing a scan so OCR actually works.

Why image quality dominates OCR accuracy

The engine has to decide, for every mark on the page, what letter is this? Everything that blurs the boundary between ink and paper makes that decision harder:

• Uneven lighting / shadows. A gradient across the page means the same letter is dark in one corner and pale in another — the engine has no consistent notion of what "ink" looks like.
• Low contrast. Grey text on a grey background gives weak edges, and edges are what letterform matching depends on.
• Softness / blur. Fuzzy letters merge into each other. This is what turns rn into m.
• Skew and rotation. OCR engines read along horizontal lines. A sideways page is catastrophic; even a few degrees of tilt hurts.
• Speckle and noise. Scanner dust and paper texture get read as punctuation, producing stray characters.

Enhancement fixes the first three directly, and it's why the same document can go from unusable to near-perfect recognition with no change to the OCR settings at all.

The order that matters

This is the crux of the whole guide, and it's easy to get backwards:

1. Rotate — get the page upright first. Rotate PDF.
2. Enhance — flatten the lighting, whiten the background, sharpen the text. Enhance PDF.
3. OCR — now run the recognition. OCR PDF.
4. Compress — last, if you need a smaller file.

Never compress before OCR. Aggressive compression softens and mangles exactly the letterforms the engine is trying to read. And never OCR a sideways page — fix the orientation first.

The steps

1. Fix the orientation. If pages are sideways or upside-down, run them through Rotate PDF first. OCR reads left-to-right along horizontal lines; a rotated page confuses it completely.

2. Open Enhance PDF and drop your scan in.

3. Push Whiten Background up. For OCR prep, a clean, uniformly white background is your friend — it gives the engine an unambiguous distinction between paper and ink. The default (75) is a good starting point; go higher if any shadow remains.

4. Raise Contrast. Strong separation between text and background is exactly what letterform matching needs. This is arguably the single most valuable setting for OCR accuracy.

5. Add Sharpness — but carefully. A moderate bump crisps the letter edges and genuinely helps. Too much creates a halo of artefacts around each character, which the engine can read as extra marks. Use the preview.

6. Click Enhance PDF and download.

7. Now run OCR PDF on the enhanced file.

The one exception: faded documents

Everything above assumes normal, well-inked text. If your document is faded — pencil, an old carbon copy, a dying thermal receipt — pushing Whiten Background up will delete the very text you want OCR to read.

For those, invert the advice: keep whitening low and lean on contrast. Our guide on enhancing a faded document covers the trade-off.

Common questions

Does enhancing really improve OCR accuracy that much?

Yes, and it's the highest-leverage thing you can do. OCR engines are sensitive to contrast and shadow in a way that human eyes aren't — a page you can read perfectly well may still be genuinely hard for the engine. The difference between OCR on a raw phone photo and on the same photo enhanced is often dramatic.

Should I compress before or after OCR?

After. Always. Compression works by degrading images, and degraded letterforms are precisely what breaks recognition. Enhance → OCR → compress. See compressing a scanned PDF.

My page is tilted a few degrees. Does that matter?

Yes — OCR reads along horizontal lines, and even a small tilt reduces accuracy. Rotation fixes 90° problems but can't correct a 3° skew. For a badly skewed page, the honest fix is to re-scan it squarely.

Does the grayscale output hurt OCR?

No — quite the opposite. OCR doesn't use colour; it works on the shapes of letters. A clean black-on-white grayscale page is the ideal input. The colour information in a scan is noise as far as the engine is concerned.

I already OCR'd it and the text is garbage. Can I redo it?

Yes. Go back to the original scan (not the OCR'd output), enhance it properly, and OCR again. Running OCR a second time on a bad text layer doesn't improve it — you have to fix the image.

What if the scan is already clean?

Then you may not need to enhance at all. A crisp, well-lit, flat-white scan from a proper flatbed is already what OCR wants. Enhancement is for rescuing the imperfect ones — mostly phone photos and old scans.

Wrap-up

The order is the whole lesson:

1. Rotate — get it upright.
2. Enhance — whiten the background, raise contrast, sharpen modestly.
3. OCR — now recognition has something clean to work with.
4. Compress — last, never before OCR.

OCR accuracy is mostly an image-quality problem wearing a software costume. Clean the image and the engine does the rest. For the full scanning workflow, see scanning a document with your phone.