Guide
Straightening crooked scans before you share or OCR them
Paper rarely goes through a scanner perfectly straight. A page fed half a degree off is barely noticeable; three degrees makes a document look careless, makes it harder to read, and can cut OCR accuracy noticeably. Rotating each page by eye in an image editor is slow and imprecise. This tool measures the tilt of every page from its lines of text and straightens the pages that need it.
How the tilt is measured
Lines of text are the best guide to how a page sits. The tool renders each page, takes the dark pixels and counts them row by row along a range of slightly tilted angles. When the angle matches the text lines, the counts form sharp peaks (lines) and valleys (gaps between lines). The angle with the sharpest pattern is the tilt, measured to a few hundredths of a degree.
The same measure tells how clear the result is. A page of text gives a strong, unambiguous peak. A photo, a blank page or a page with only a few words gives a weak one; those pages are reported as having no clear text lines and are left alone unless you type an angle.
Reading the table
Each page gets a row:
- Detected tilt: for example +1.84°. Positive means the text runs downhill to the right, negative uphill.
- Correction: what will be applied. It starts at the detected angle for clearly tilted pages and at 0 for straight or unclear ones.
Change any correction by typing a value or using the arrows in 0.1° steps. Set it to 0 to leave a page as it is. The button tells you how many pages will be straightened.
What happens to the pages
Pages that are straightened are rendered at the resolution you choose, turned, and saved back at their original paper size, with white filling the corners that the rotation exposes. Because the page becomes an image, any text layer it had is not kept; for scans that is normally no loss.
Pages with no correction are copied into the new file untouched: same content, same quality, same selectable text. In a mixed document only the crooked scans change.
Choosing the resolution
- 300 DPI for documents you will OCR, archive or print.
- 200 DPI, the default, for sharing on screen and by email.
- 150 DPI to keep the file small.
Deskew first, then OCR
Text recognition reads lines. Straight lines are recognised more accurately and the recognised words line up better with the image. Straighten the scan here, then run OCR PDF on the result.
A typical clean-up order for scans
- Fix page order with Reverse PDF Pages and turn sideways pages with Rotate PDF.
- Straighten with this tool.
- Trim dark scanner edges with Crop PDF.
- Run OCR, then Compress PDF if the file is large.
The whole process runs in your browser, so scanned contracts and records never leave your computer.
A worked example: a scanned lease
A 12-page lease came out of a sheet-fed scanner with several pages tilted.
- Open the PDF; the table shows tilts between -2.4 and +1.9 degrees on 7 pages, and 5 pages already straight.
- Page 12 is a signature page with little text and shows no clear text lines; it looks straight, so leave it at 0.
- Choose 300 DPI, because the lease will be OCR’d, and straighten 7 pages.
Checked against the current version of the Deskew PDF on .