Character recognition

Which is the best method for character recognition? I mean which method would you use for comparison? and it should work perfectly for all inputs.

I used correlation to compare between images and I am not quite satisfied with it. Give me some other best method for character recognition. What do you think about bayes classifier and neural networks?

Answer 1

[Greg Heath]

The best method depends on the particular data set.

I have had very good results with neural networks. A newsgroup search using

heath character recognition

should find my code.

Be sure to add noise to the images and characterize your performance in terms of signal-to-noise ratio.

Hope this helps.

Greg

Answer 2

[Walter Roberson]

An OCR method that "should work perfectly for all inputs" would solve the CAPTCHA problem. Some very good researchers have been working on that for quite a number of years.

I am, I find, confused about how you would apply such a "perfectly" working system against Chinese and (to a lesser extent) Japanese, and other languages in which the numbers are also words.

You will, I trust, allow archeologists to use your system so that they can automatically decipher all of those fragments they have in languages that have not been translated yet. I imagine that archivists would be very happy that you will be able to recover pages that have been largely washed away or had ink spilled on them, or been eaten by insects or mold. And SETI would, I am sure, be very very happy to have the system applied to potential signals received from outer space.

Communication with people with extremely poor muscle control. Communication with animals...

What are you going to do with your second billion dollars in profits?