How to straighten a crooked scan before OCR

Read by Zaweba straightens crooked scans for you. Its deskew step measures how far the page is tilted and corrects it, over a range of plus or minus 15 degrees, before the text is read. You open the scan, Read does the straightening as part of its clean-up, and you get both the level page and its text.

A page that slipped on the scanner glass, or a sheet that went through a feeder at a slant, looks harmless to the eye. OCR engines are fussier. They expect lines of text to run straight across, and even a few degrees of tilt can make them split lines, merge words or lose the ends of rows.

How precise the straightening is

Read measures the angle in very fine steps, down to 0.05 degrees within its 15 degree range. In testing, a scan that was 4.5 degrees crooked was measured to within about 0.2 degrees. That is close enough that the corrected page looks level to the eye and the lines of text run straight for the engine.

Deskew is one of several clean-up steps, alongside turning pages upright, removing shadows, lifting faint print and finding the page in a photo. Each one acts only when it is needed, so a scan that is already straight is left alone rather than nudged.

The steps in order

Start Read by Zaweba and drop the crooked scan on the window. Pictures, multi-page TIFFs and scanned PDFs all work, and you can scan straight into Read if your scanner uses WIA. A PDF is turned into pictures at 300 DPI, one per page, before the clean-up runs.

Have a look at the Document tab. It tells you which clean-up steps were applied, so you can see that the page was straightened. The review then shows the straightened picture next to the text. If Tesseract did the reading, words it is unsure of are shaded amber, and F8 takes you from one to the next so you can correct them.

Save the result as you need it. Plain text and .docx carry the words. A searchable PDF uses the cleaned-up, straightened picture as the page with the words laid over it as invisible text, so the file you keep is neater than the scan you started with.

What deskew can't do

Fifteen degrees covers ordinary slips, not a page put in at a wild angle. If the page is badly askew, scan it again with it closer to straight, or let Read handle it as a photo. Deskew also turns the page as a whole. A photo taken at an angle, where the top of the page is narrower than the bottom, needs flattening rather than rotating, and that is the job of the find the page and flatten step on the Clean-up tab.

Pages that are on their side or upside down are a separate case, handled by Read's orientation check. You don't need to do anything different; the two steps run together.

Scanning so there is less to fix

A little care at the scanner helps. Push the page against the edge of the glass, close the lid gently, and for a feeder, square up the stack before you start. Read's own Scan dialog lets you pick 150 to 600 DPI, greyscale, colour or black and white, and the feeder if there is one. For ordinary printed text, a middling resolution in greyscale is a sensible starting point.

For a box of old papers, use the Batch tab. Every page is straightened and read in the background while you carry on with something else, all on your own PC with Read by Zaweba.