Google Docs and various other services also use the same Tesseract OCR engine. You can also use FreeOCR online by uploading PDF files to. HP made it open source in 2005, and Google now maintains the source code. FreeOCR is based on the widely used Tesseract OCR engine, which was originally developed by Hewlett-Packard in England in the 1980s. If you want to try a few pages as an experiment, then you can download FreeOCR for Windows, though be careful not to install any crapware that may be included. If you have 250 words on a page and the program miraculously gets 90% of them right, you will still have to correct 25 words. When it is processing a document, it will present you with words it doesn’t recognise, so that you can tell it what they are. Like most if not all the programs in this field, SoftWriting has to be trained to recognise your handwriting. But it also says it is designed “for recognising non-connected handwriting and machine-printed text” (their emphasis) so I wouldn’t bet on it reading your handwritten notes. There’s also CharacTell’s SoftWriting ($49.95), which the company says is for students taking notes in class and professionals taking notes in meetings. Both companies offer free trial versions so you can test them before you splash out. It’s generally agreed that the best OCR programs are Abbyy FineReader (£99) and Nuance’s Omni (£79.99) and Ultimate (£169.99), though neither is suitable for cursive handwriting recognition. However, all this has more to do with keyboard replacement strategies than with OCR. I’ve had some success with this approach, starting more than a decade ago with Microsoft OneNote (which can also record your voice in sync) running on Windows XP Tablet Edition, and more recently with a Livescribe Echo digital pen and MyScript software.

That enables you to train the software to recognise your input, while the software also trains you to write characters in ways that it can understand. In my experience, you can only get handwriting recognition to work well enough by doing it in real time. OCR works best with high-quality printed materials and worst of all with handwriting, so you’re not starting from the best position. The same software wouldn’t have the domain expertise to cope with a Russian-speaking coroner who liked to include Sanskrit quotations in his handwritten autopsies. For example, it’s possible to recognise the English names for numbers and the names of major UK cities, especially if you can get people to write each letter in its own little box. But in practice, it works best when dealing with restricted inputs and/or limited domains.

Scan a document into word