Deskew Before OCR: Getting the Angle Right

Deskew is usually run on the way to OCR, not for its own sake. The setting you choose decides whether a page is straightened or quietly left alone.

Why this comes up

Feeder scans lean by a fraction of a degree from page to page, while a phone photo or a photocopy can lean several degrees in the same direction. One control cannot correct both well.

Check before you run it

• Use automatic detection for a stack with varying tilt. Leaving the angle blank measures each page separately, which is what a document feeder produces. A fixed angle applies the same correction everywhere and will over-rotate the pages that were already straight.
• Protect pages that are meant to be angled. The strongly-skewed option leaves small deliberate tilts alone, and the pages box limits the run to the sheets that need it. Covers, fold-outs and landscape inserts often belong outside the correction.
• Check the first and last page before OCR. Automatic detection reads the ink, so a page with a large graphic or almost no text can be misread. Compare the top edge and the text baseline, then re-run just those pages with an angle.

When to reach for it

• A batch goes straight from the scanner into OCR. Deskew the scan first so recognition sees a level baseline; straightening afterwards means checking the text layer again.
• One section of a document leans while the rest is square. Limit the page range to the leaning section instead of rotating the whole file by one guessed angle.

One question we get

When do I need deskew?

Scans that came out a degree or two tilted, usually from a feeder. The tool measures the lean and straightens each page — automatic, or page by page with a rotation dial.

Where it fits

The tools people usually open next to Deskew:

• Grayscale — Make a PDF text and images grayscale
• Bates Numbering — Bates stamp multiple files at once.
• Extract Images — Save each image from the PDF as a separate file

Every tool works on a copy, so the original file is never touched. Uploads are stored under an unguessable id and deleted automatically two hours after you last open them; nothing is listed, shared or indexed. More on the security page.