How the contrast boost in Read helps with faint print

Faint print is one of the quiet reasons text recognition goes wrong. Old photocopies, worn receipts, pages printed when the toner was running low and scans made with the brightness too high all have the same problem: the letters are only a little darker than the paper. Read by Zaweba's contrast boost lifts them before reading, so the engine has something clearer to work with.

It's one of the clean-up steps Read runs on a picture before the text is read. Like the others, it acts only when it's needed.

What a percentile stretch does

The boost is a percentile stretch. Read looks at how dark and how light the pixels on the page are, ignoring a small share at each extreme, such as a stray black speck or a bright reflection. It then stretches what's left so the darkest of the rest becomes properly dark and the lightest becomes properly light.

On a washed-out page, the text might sit in pale grey on slightly paler grey. After the stretch, the same text is near-black on near-white. Nothing is drawn in or guessed; the picture's own differences are simply made bigger. Ignoring the extremes is what makes it reliable, because one dark mark can't stop the rest of the page from being stretched.

Where it fits with the other clean-up steps

Read's Clean-up tab covers several jobs. It finds the page in a photo and flattens its perspective, removes shadows by estimating the background and dividing it out, turns sideways or upside-down pages upright and straightens crooked ones. The contrast boost has a narrower job: making faint print stand out from the paper.

The steps complement each other. Shadow removal deals with light that varies across the page, such as a dark corner where your hand blocked the lamp. The boost deals with print that is pale everywhere. A phone photo of an old photocopy can need both, and Read can apply both.

Each step acts only when needed, and the Document tab says what Read did. That's useful when a result surprises you, because you can see whether the boost was applied.

Knowing when it won't help

Contrast can only make visible what is already in the picture. If the print is so faint that the scanner or camera didn't capture it, there's nothing to stretch. Likewise, a picture taken at low resolution has lost the shapes of the letters, and making it darker won't bring them back.

In those cases, go back to the source if you can. Rescan at a higher resolution, which Read's scan dialog offers from 150 to 600 DPI; try greyscale rather than black and white, which can throw faint print away; or retake the photo in better light.

Checking what came out

After reading, the side-by-side review shows the cleaned-up picture beside the text. With the Tesseract engine, words under 70 per cent confidence are shaded amber, and faint print often accounts for a cluster of them. Press F8 to jump from one to the next and fix anything wrong. You can switch on a box round every word to see how the engine split the page.

If you're working through a whole box of faded paperwork, the Batch tab runs the same clean-up on every page in the background and reports how many words each file has to check. Read by Zaweba does all of this on your own PC.