Account

What new tool do you want?

PDF Tools

Straighten & Clean Scanned PDF (Deskew)

Automatically straighten crooked scanned pages and whiten the paper background — fix tilted scans, remove gray noise and shrink the file with grayscale. Runs in your browser.

How to use

  1. 1Upload a scanned PDF.
  2. 2Choose quality, background cleaning and the tilt threshold.
  3. 3Download the straightened PDF — the note shows how much each page was rotated.

Frequently asked questions

How is the tilt detected?

Text lines are found automatically: the page is tested at angles up to ±6° and the angle where lines line up best is used, to within 0.05°.

Will the text stay selectable?

Pages are re-created as images. Run OCR afterwards to make the text searchable again.

Is my PDF uploaded?

No — it's processed on your device.

How the tilt angle is actually found

The detection works by rendering each page to a grayscale canvas and testing a range of rotation angles (up to roughly ±6 degrees, per the tool's own FAQ) to find the one where the page's text lines align most consistently into straight horizontal rows — the logic being that correctly oriented text produces sharply defined horizontal bands, while a crooked page produces blurrier, misaligned ones. This only works because most scanned documents are text-dominant; a page that's mostly a photograph or a technical diagram with no clear horizontal text lines gives the detector little to measure, which is reflected in the confidence score the tool computes alongside the angle.

Why a detected tilt isn't always corrected

Two independent gates decide whether a page actually gets rotated: the detector's own confidence must exceed a minimum threshold (so a low-confidence guess on an image-heavy or blank page doesn't introduce an unwanted rotation), and the detected angle must exceed whatever minimum you set in Straighten If Tilted More Than. Setting that threshold to 1 degree means pages tilted by a barely-perceptible 0.6 degrees are deliberately left alone — useful because over-correcting a near-straight page based on a noisy, low-confidence angle estimate can sometimes make it look worse, not better.

The trade-off every mode shares: pages become images

Regardless of settings, every page is re-rendered as a JPEG at your chosen DPI (150, 200 or 300) with the straightening rotation and any background cleaning baked directly into the pixels — there is no mode that straightens a page while preserving its original selectable text, because rotating a scanned image and rotating a true text layer are fundamentally different operations, and this tool only works with the rasterized page. The tool's own FAQ states this plainly: run OCR PDF afterward if the straightened result needs to be searchable again, since straightening by itself only fixes the visual tilt.

Example: straightening a 30-page batch-scanned archive

A box of old paper records gets fed through a sheet-fed scanner, and about a third of the resulting 30-page PDF comes out tilted by anywhere from 0.5 to 3 degrees because the pages weren't perfectly aligned going through the feeder. Running Deskew with the default 0.2-degree threshold and Light background cleaning catches and corrects essentially all of the tilted pages while leaving the handful of already-straight ones untouched (since their detected angle falls under the threshold), and the result note lists exactly which pages were rotated and by how much — e.g. 'Straightened 11 of 30 pages (+1.2°, -0.8°, +2.1°…)' — so you can spot-check the corrections before relying on the batch.

Related tools

More: PDF Tools