# Prepare Poor-Quality Scans Before OCR: Blur, Skew, and Shadows

Diagnose common scan defects, use external preprocessing carefully, compare the same sample before and after, and recognize information that cleanup cannot restore.

Before processing a difficult scan, identify the defect. Tilted text, a sideways page, blur, dark binding shadows, and missing margins need different responses. Applying every cleanup option can erase useful marks as easily as it removes noise.

Keep the original untouched. Test changes on a copy of a few representative pages before applying them to the whole book.

## Separate defects you can correct from missing information

| Defect                       | What to inspect                         | Useful next step                                  |
| ---------------------------- | --------------------------------------- | ------------------------------------------------- |
| A small tilt                 | Lines slope but remain sharp            | Test deskewing                                    |
| Sideways or upside-down page | Reading orientation is wrong            | Correct rotation                                  |
| Binding shadow               | Small text disappears toward the gutter | Prefer a better capture when possible             |
| Blur or motion               | Letter strokes merge                    | Rescan or obtain a clearer source                 |
| Cropped margins              | Characters or note markers are absent   | Recover a complete original page                  |
| Speckles or dark borders     | Noise competes with small type          | Test conservative cleanup without erasing content |

Increasing the size of an already blurry image does not restore missing character detail. A plausible word guessed from context is not the same as recovering the source. Mark genuinely unreadable material instead of treating fluent OCR or translation as proof.

## Use external preprocessing deliberately

These are external preparation options, not PDF Craft image-editing controls. Adobe documents its [scan enhancement tools](https://helpx.adobe.com/acrobat/desktop/create-documents/scan-documents-to-pdfs/enhance-scans.html) under Scan & OCR. Follow the current instructions for your version, save a working copy, and inspect the result before replacing anything.

For a command-line workflow, [OCRmyPDF's cookbook](https://ocrmypdf.readthedocs.io/en/stable/cookbook.html) distinguishes rotation from deskewing. For an appropriately prepared test file, its documented options include:

```sh
ocrmypdf --deskew --rotate-pages input.pdf output.pdf
```

This example writes a separate output; it is not a command you must run before every PDF Craft task. Install and configure the external tool according to its documentation, select appropriate recognition languages where required, and read warnings for the actual source. Do not add destructive cleanup options simply because an unrelated tutorial uses them.

[Tesseract's quality guidance](https://tesseract-ocr.github.io/tessdoc/ImproveQuality.html) explains why input quality and skew affect recognition. Its advice concerns that OCR engine, not a promise that one universal setting solves every book.

## Compare the same pages before and after

Choose a normal page and a difficult page with small type, notes, or a table. Keep their original locations. Apply one meaningful adjustment at a time so you can tell what helped and what damaged the image.

Compare both visual content and recognized text. Check punctuation, decimal points, minus signs, superscripts, and thin table lines. Strong contrast can make a page look cleaner while removing faint strokes that matter.

Also inspect margins and the binding area. A crop that removes a black border may remove a footnote marker too. A cleanup pass that improves body text may harm illustrations or handwriting. Do not judge success only by the most prominent paragraph.

Record the settings and unresolved defects. A small sample helps estimate the review burden; it does not establish a whole-book accuracy percentage.

## Choose the next output only after the source is usable

Use EPUB when the aim is continued reading with adjustable text. Use Markdown when the aim is editing or research notes. Keep the PDF for original page appearance and source checks.

In the [PDF workspace](https://pdfcraft.ai/en/pdf-craft/), upload the prepared PDF and choose the actual output and translation settings. Translation is optional for EPUB and Markdown, while PDF output requires translation. The website URL does not make those choices for you.

New users receive trial credits, and processing is charged by usage. Review current limits and [Pricing](https://pdfcraft.ai/en/pricing/) before submitting a large collection. PDF Craft does not promise to restore every damaged character or reproduce all complex layouts without review.

## Preserve uncertainty in the final check

After download, return to the same difficult pages. Compare paragraph order, small notes, formulas, and the text near the gutter. If translation is involved, check the original image when a sentence seems odd; the error may have started with recognition.

The [scanned PDF translation guide](https://pdfcraft.ai/en/blog/translate-scanned-pdf/) covers downstream review. The [OCR and translation guide](https://pdfcraft.ai/en/blog/ocr-and-translate-workflow/) explains why recognition and meaning need separate checks. Keep the original and the prepared copy together so later corrections remain traceable.