How to Straighten a Scanned PDF (Deskew Crooked Pages, Free)

You photographed a contract on your desk, or fed a stack through a tired office scanner, and every page leans a few degrees — enough to look sloppy, strain the eyes, and noticeably hurt OCR accuracy. The fix is deskewing: detecting each page's tilt and rotating it back to level. Here's how to do it in a browser, what the correction actually does to your file, and the one trade-off you should know before you start.

Where the tilt comes from

• Sheet-fed scanners. Rollers grab the paper slightly unevenly, so pages drift 1–3° — randomly, each in its own direction.
• Phone photos. Holding a camera perfectly square over a desk is nearly impossible freehand; a 2–5° lean is normal.
• Flatbed haste. The original wasn't flush against the guide rail.
• Rebound copies. Photocopies of photocopies accumulate small rotations over generations.

Note the difference from a sideways page: if a page is off by 90° or 180°, that's a rotation problem — fix it with Rotate. Deskew handles the small, awkward angles in between that rotation tools can't touch.

How to straighten a scanned PDF — step by step

🔒 pdfty.com/en/tools/deskew

What deskewing actually does — and the trade-off

Deskew renders each page to an image, finds the skew angle (typically by following the direction of the text baselines), rotates the image back by that angle, and writes the corrected image into a new PDF page.

That middle word — image — is the honest part most tools skip over:

Why per-page detection matters: sheet-fed scanners skew every page by a different random angle. A single global rotation can't fix that — one correct angle per page can, which is exactly what automatic detection gives you.

Is a couple of degrees worth fixing? For a file you'll read once, maybe not. But for anything you'll share, archive, print or OCR, yes: skewed pages print with uneven margins, look careless in exhibits and reports, and degrade text recognition. Since the fix takes under a minute and applies to the whole file at once, there's little reason to live with the lean.

Straightening locally instead

Free offline routes, with their real trade-offs:

• Rescan properly. The cheapest fix when you still have the paper: square the stack, use the guide rails, or use a scanning app with live edge detection. No software repairs beat a straight scan.
• NAPS2 (Windows/Mac/Linux, free). A scanning app with a built-in deskew filter — great if you control the scanner, no help for PDFs someone else sent you.
• ImageMagick (command line). magick convert -deskew 40% in.png out.png straightens individual images; for a PDF you'd burst it to images, deskew each, and rebuild — scriptable but fiddly.
• ScanTailor (desktop, free). Post-processing for book scans with excellent manual deskew control. Powerful, but a learning curve and an image-in/image-out workflow.

All of these rasterize or work on images too — rotation by an arbitrary angle fundamentally re-renders the page. No deskew tool keeps a selectable text layer through the rotation; the ones that appear to are running OCR afterwards.

Practical gotchas

• Deskew before OCR, always. OCR engines misread slanted baselines; a few degrees of skew measurably increases recognition errors. The right pipeline is deskew → OCR — see our guide to free OCR for PDFs.
• Edge wedges. Rotating a page leaves small triangular areas at the corners where there was no source image; they're filled with the background color. On clean white scans you'll never notice.
• Severely warped pages are a different problem. Deskew fixes rotation — a flat page at an angle. A phone photo with perspective distortion or curved book-spine text needs dewarping, which is a heavier operation; expect only partial improvement.
• File size can change. Re-encoding page images can grow the file, especially from aggressively compressed originals. If the output is heavy, follow with Compress — see how to make a PDF smaller.

Frequently asked questions

Will the text still be selectable after deskewing?

No — this is the one real trade-off. Pages are rotated as images, so any existing text layer is lost. Run OCR on the straightened file to make it searchable again.

How big an angle can it fix?

The typical scanner/phone range — a few degrees either way — is the sweet spot. A page that's fully sideways (90°/180°) should be fixed with Rotate first.

Does every page get the same correction?

No — each page is measured and corrected independently, which is exactly what sheet-fed scans need, since every page skews differently.

Will it fix perspective distortion from phone photos?

Only the rotational part. Keystone/perspective distortion (the page looks like a trapezoid) is a different correction; deskew will level the text but can't un-warp it.

Can it make a blurry scan sharper?

No. Deskew rotates pixels; it doesn't invent detail. Quality in equals quality out — minus the tilt.

What are the free limits?

20 MB and 50 pages per file, 10 operations a day, no sign-up, no watermark. Files are deleted within 1 hour. Pro ($9/mo) lifts the limits.

Should I deskew before or after removing blank pages?

Order barely matters for quality, but deskewing fewer pages is faster — clean the file first with Delete Blank Pages, then straighten.

Why is my deskewed file bigger than the original?

The rotated page images are re-encoded, sometimes less compactly than the original scanner's compression. Compress usually recovers the difference.