2

What would be the best/most accurate way to read human writing from a scanned document and convert it into text?

2
  • 1
    I think your question is too broad - any hand-writing OCR software would do. However, don't expect a very good quality from any of them; recognizing hand-written text is often a hard task even with Natural Intelligence ;-), it is yet more hard by AI.
    – peterh
    Dec 9, 2019 at 18:10
  • According to this, the only OCR engines that recognize handwriting are Azure and Abbyy. Currently Tesseract and Transym only recognize machine print.
    – osucowboy
    Feb 19, 2021 at 15:24

3 Answers 3

2

If you have a lot of handwritten text from a single author you can train some OCR systems on that text and possibly get a reasonable accuracy rate but do not expect anything much above 80% unless the author has a near machine look script - after all we all have problems reading some peoples hand written text sometimes even our own.

Among the OCR systems that can be trained one to consider is Tesseract OCR which as of version 4 includes a Neural Net based recognition system, details on training it can be found here.

Tesseract OCR is:

  • Free, Gratis
  • Free, Open Source
  • Apache 2.0 Licence
  • Cross Platform
  • Mature it has been about since at least 1985
  • Actively Maintained as of Dec 2019
  • Has UTF-8 Support
  • Can process text in over 100 languages "out of the box"
  • Can process Right to Left Language text
  • Supports multiple output formats.

It does not have it's own GUI but there are some 3rd Party GUIs available.

Note: to get better OCR results, you'll need to improve the quality of the image you are giving Tesseract.

0

Of all the OCR softwares, I have found myself using Microsoft's OneNote a lot nowadays. You can take a picture in the app itself or paste a picture in the notes, and it will generate OCR text, though it takes a couple of minutes. It's present on both mobile and PC so has flexibility.

0

For recognizing handwritten text, what you are looking for is ICR, or Intelligent Character Recognition.

OCR, or Optical Character Recognition, is used for printed text that remains more or less consistent from letter to letter in terms of shape due to using a set font. For handwritten text however, there is no standard font and you have to analyze and anticipate the intended shape from a bitmap which is usually a bit harder computationally.

You may want to try out the LEADTOOLS ICR libraries (www.leadtools.com) as they have released an ICR engine in the latest version which is cross platform compatible through .NET Core.

It does not require lengthy training as machine learning libraries would, and can generate decent output even before finetuning, as long as the input image is fairly clean and has high contrast. They also have image processing functions to clean up documents and images as well which aids in recognition.

Below you can see an example with one of their demos.

Disclaimer: I work for LEAD Technologies who develops this toolkit.

Example screenshot of LEADTOOLS ICR libraries result

2
  • The provided image does not look like handwritten text. It looks like a script type PC font. Jan 28, 2021 at 22:57
  • I changed out the image for one of the Forms demos. The Forms libraries can also match a scanned image against a masterform template and extract various types of information, such as OCR and ICR, from predefined field locations.
    – JClark742
    Feb 1, 2021 at 14:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.