Deskew / Straighten Scans - Free Online Browser Tool
Measure how far each scanned page is tilted and rotate it back to straight, automatically.
A stack of photocopies that went through a lid that did not quite close. Every page leans a couple of degrees, they are not all leaning the same way, and a document you cannot stack square or feed through a guillotine is a document that causes trouble downstream. You can correct each page by hand in a PDF editor, which is slow, and more importantly it is guesswork - nobody can see two degrees by eye, so the second page comes out different from the first. This measures instead. For each page it finds where the dark pixels sit along every row, then works out the angle that would line all that text back up into vertical columns. That is the tilt the page was fed in at, and undoing it is the correction. A page fed in at four degrees comes out at four degrees, and a page fed in square comes out at exactly zero rather than at some small angle that looks like a rounding error and would soften the page anyway. Now the part worth reading before you use it. The result is a picture of each page rather than a page. That is not a shortcut, it is the format talking: a PDF page angle can only be 0, 90, 180 or 270 degrees, so there is no way to record a four-degree tilt in a page's rotation at all. The only way to show a page leaning four degrees is to draw it leaning four degrees. What you get back looks identical to what you put in - which is rather the point, since this is a scan - but the text layer does not come with it, so the result is no longer searchable or selectable. If the original had a text layer you care about, keep the original. It is also honest about what it cannot measure. A photograph, a blank sheet, or a cover with one line of text on it has too little on the page to tell a two-degree tilt from noise, and rotating on a guess there would be worse than doing nothing. Those pages are listed as unmeasurable rather than silently adjusted, and you can ask it to stop instead of skipping them if you would rather know. Pages that are already straight are left alone, because re-rendering a good page at a fraction of a degree loses a little sharpness for no gain at all. If you want to turn a phone photo into a clean page in the first place, <a href="https://www.allpdf.tech/images-to-pdf/camera-scan-pdf" class="seo-link">Camera Scan to PDF</a> does the whole job including the squaring-up. This one is for the stack that is already a PDF and already needs tilting fixed. If you want the tilt on one page turned a deliberate 90 degrees rather than corrected, Rotate Selected Pages Only is the tool for that and it does not rasterise anything. Everything runs in this tab; the file is never uploaded.
This deskew / straighten scans utility runs completely inside your web browser. Your document is opened locally and processed on your own device using JavaScript and WebAssembly, so your file is never uploaded to any server. There is no signup, no watermark and no limit on how many documents you can handle.
- 100% client-side - files never leave your device
- Instant processing, no upload wait
- Works on desktop, tablet and mobile
- Free forever, no account required
How to Use the Deskew / Straighten Scans Tool
Five steps, and the whole thing runs locally on your own device.
-
Add the scanned PDF
One file. Each page is rendered small to measure it, then rendered again at full size to be redrawn - the measurement does not need the detail, the output does.
-
Choose how strict to be about pages that are already straight
By default anything under four tenths of a degree is left exactly as it is, since re-rendering it would cost sharpness and gain nothing.
-
Decide what to do with pages it cannot measure
Leave them as they are and carry on, or stop and have it name them - useful when you need to know that a cover came out untouched.
-
Narrow to certain pages if you need to
A range like 3-12, or leave it blank for the whole document.
-
Save
One straightened PDF plus a per-page report: the tilt found on each, which were corrected, which were already straight, and which could not be measured at all.
Frequently Asked Questions
Why does the text stop being searchable?
Because each page comes back as an image. A PDF page can only record an angle of 0, 90, 180 or 270 degrees, so a four-degree tilt cannot be stored in the page itself - the page has to be redrawn at that angle, and a redrawn page is a picture. For a scan this usually costs nothing, since a scan has no text layer to begin with; it matters if your original was OCR-ed and you want to keep that layer.
How accurate is the measurement?
Within about a tenth of a degree on a page with ordinary body text, tested against pages rotated by known amounts from two to nine degrees in both directions. It reads a page that is already straight as exactly zero, which is the case a simpler approach gets wrong by confidently reporting several degrees.
What happens to a page it cannot measure?
It is left alone and listed as unmeasurable, rather than being rotated on a guess. A photograph, a blank sheet or a cover with a single line on it has too little on it to tell a small tilt from noise, and a wrong rotation is much worse than no rotation. You can also ask it to stop and name those pages instead of carrying on.
Will it straighten a document that is already straight?
No, and that is deliberate. Pages measuring under four tenths of a degree are left untouched, because re-rendering a good page costs a little sharpness and gains nothing. If you want every page corrected regardless of how small the tilt is, that setting can be changed - but on a straight document it will only make the pages slightly worse.
Is this the same as rotating a page?
No. Rotate Selected Pages Only turns a page by an angle you choose and keeps everything intact, because a quarter turn is something a PDF page can record. This measures what the scanner did and undoes it, which is a correction rather than an instruction - and that difference is exactly why this one has to redraw the page.