What Read by Zaweba's Help tab says, including its limits

Every text recognition app has things it does badly, and the useful ones say so. Read by Zaweba's Help tab covers getting started, how its two engines work, what it does with your files, and its honest limits: handwriting, low-resolution pictures and decorative fonts aren't read reliably.

Knowing those limits before you start saves time. You'll know which jobs to give Read, and which need a different approach.

The basic loop

The Help tab begins with the basics. Get a picture in: open a file, drop files or folders on the window, paste a picture or files with Ctrl+V, open a scanned PDF or scan from a WIA scanner. Read cleans up the picture if it needs it, reads it and shows the page beside its text. Check the text, correct anything wrong, then copy it or save it as text, .docx, a searchable PDF or CSV.

That's the whole loop. Everything else in Read, from batches to the screen grab, is a faster way to go round it.

Engines in plain words

Read has two engines. Windows OCR is built into Windows, is fast, reads one language at a time from those installed in Windows, and gives no word confidences. Tesseract 5.0 comes with Read, analyses layout, gives a confidence for every word and supports 40+ languages, several at once.

In Auto, Read picks Windows OCR for quick single pictures when Windows has the language, and Tesseract for columns, searchable PDFs, tables, multi-page documents and languages Windows lacks. The engine used and the reason appear above the text.

Privacy

The Help tab is plain about privacy too. Nothing leaves the PC except language packs you choose to download, and the update check is off until you turn it on.

The honest limits

Handwriting is the big one. Both engines are built for printed text. Neat block capitals may come out partly right, but joined-up writing usually won't, so don't rely on Read for handwritten notes.

Low-resolution pictures are next. Text recognition needs to see the shapes of the letters. A small thumbnail or a photo of a page taken from across the room hasn't captured them, and no clean-up can put back detail that isn't there. Rescan or retake the picture closer, or at a higher resolution; Read's scan dialog goes from 150 to 600 DPI.

Decorative fonts are the third. Script faces, heavy display lettering and ornate headings confuse engines trained on ordinary type. Body text in the same document will usually be fine.

Working within the limits

Most poor results come from the picture rather than the engine. Take photos square on in even light, fill the frame with the page, and keep the camera steady. Then let Read's clean-up deal with the rest: finding and flattening the page, removing shadows, turning it upright, straightening it and lifting faint print.

Use the review. With Tesseract, words under 70 per cent confidence are shaded amber, and F8 jumps to the next one. A page with a lot of amber is a sign the picture is near a limit. If Windows OCR read it, Read > Read again with Tesseract gives you the shading.

Read by Zaweba won't pretend to be perfect, and the Help tab is there so you know where the edges are.