Handwriting recognition is an important but very complicated domain of Computer Sciences. Computers nowadays do a quite fair job even if there is room for improvement in the future.

I am wondering if the result is language-dependant: 1) Are there languages that are more easily analyzed by handwriting recognition programs? Why? 2) Quality of recognition depends probably on individuals' handwriting. If there are difference of quality of recognitions between language, is this difference significantly larger or smaller than the individual variation?

Concerning the first question, my intuition is that best performances are obtained with languages with a lot of writers like English (more resources to analyze, more motivation, more money,...) or with a particular alphabet/language structure (Korean and its simple alphabet ㄱㄴㄷㄹㅁㅂㅅㅏㅓㅜ seems to be a good candidate).

Edit: I am not sure if the relevant context is language or alphabet. As Tom pointed out in comments, different languages with the same alphabet may behave differently in language recognization. For example, I think that a basic strategy when some letters of a word are identified is comparison with a dictionary. German, with its possibility of creating new words by concatenation, makes this strategy (slightly?) less useful than English. Language structure like genders may also help.

My motivation is to understand some ideas of OCR, so at the end of the day, I'll accept a good answer, even if it is focused about alphabet rather than language.

  • 2
    $\begingroup$ Welcome to Computer Science Stack Exchange! I made a very slight edit to your title to make it clearer that you're asking which human languages would be best to try to OCR, rather than which programming languages one should use to write OCR software. (That's completely clear from the text of the question but, when I saw your title, I thought it was yet another off-topic programming question.) $\endgroup$ – David Richerby Jul 29 '15 at 10:01
  • 2
    $\begingroup$ It's not quite clear if your question is about languages per se or alphabets. Of course, there are differences between the way that, say, English and French people hand-write the Roman alphabet but it seems more significant that they're both using the same alphabet than that they're writing different words in them. Also, I'd have thought that most languages would have a reasonably large amount of resources: for example, there are around 80 million Korean speakers worldwide, so there should be plenty of handwriting around. $\endgroup$ – David Richerby Jul 29 '15 at 10:06
  • 1
    $\begingroup$ @DavidRicherby Two different languages in the same alphabet might have varying difficulties to OCR, given that if some letter is unclear you may be able to infer it from the surrounding letters. A language with low entropy might be (marginally) easier to do OCR on. $\endgroup$ – Tom van der Zanden Jul 29 '15 at 11:14
  • 1
    $\begingroup$ @DavidRicherby: you did well for the title. I edited to answer about alphabets. $\endgroup$ – Taladris Jul 30 '15 at 7:16
  • $\begingroup$ I think you should separate the issues about alphabet and about languages. The difficulties are measured differently, and the accuracy is ensured in different ways in actual systems. Regarding languages, you can use entropy at character level , of course, but how much is that related to grammaticality checking, which may also help in the actual processing (wehther by statistical or algebraic techniques). $\endgroup$ – babou Jul 30 '15 at 12:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.