Discover MakerZone

MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

Learn more

Discover what MATLAB® can do for your career.

Opportunities for recent engineering grads.

Apply Today

To resolve issues starting MATLAB on Mac OS X 10.10 (Yosemite) visit: http://www.mathworks.com/matlabcentral/answers/159016

What is the best classifier for train the handwritten characters?

Asked by Minu on 13 May 2013

I am doing my project in OCR.I have tried feed forward back propagation , Perceptron etc for training .Finally use radial basis function and got more accuracy during training phase.But the problem arise during testing part.For training , i have taken 710 characters and got 97.58% performance.But during testing i have got poor results(got 5-20% accuracy). I have used image that are written by same person for testing and training phase. Now i don't know what to do... .My last option is to change my classifier. My project depends on these. I have searched online and tried for several days(nearly 1 month) to find a solution, but nothing has worked.Could anybody help me please..............

0 Comments

Minu

Products

No products are associated with this question.

2 Answers

Answer by Walter Roberson on 14 May 2013

No-one knows what the "best" classifier is. We expect that a better one will be invented within the next 50,000 years. In the meantime the best of the known classifiers very likely has performance depending on the kind of characters being analyzed. So far, no-one has found any that can read what Nature has writ in the frost on the windows.

2 Comments

Minu on 14 May 2013

I have tried neural ,svm .I dont know what'z the problem.During training i have got 95-100% for RBF.But during testing i have got very poor results.So how to resolve it.Kindly help me plz....

Jan Simon on 16 May 2013

@Minu: Reading hand-written characters is a serious problem. Note that pupils require years to learn this, although their visual system has been optimized for millions of years (and this is an understatement).

There cannot be a "best" method, because it will depend on the age and nationality of the writers, on the fact if they care for readability as on bank cheques, on the properties of the paper and on the quality of the camera, e.g. number of gray shades.

In consequence I assume, that even after years of further investigations, you will not find a "best" method, but a bunch of almost satisfying solutions.

Walter Roberson
Answer by Greg Heath on 14 May 2013

help patternnet

Make sure your target matrix contains columns of the unit matrix eye(c) where c is the number of classes.

Search for examples in NEWSGROUP and ANSWERS using

patternnet

Hope this helps.

Greg

3 Comments

Minu on 16 May 2013

I have tried patternnet and got poor results. I am unable to move forward. I am stuck in this situation for nearly one month.I dont know what to do.Kindly give me a solution.

Greg Heath on 16 May 2013

Post your code, with comments, questions and error messages.

Greg Heath on 16 May 2013

However, even if your code is perfect, you cannot be successful without a good set of features. I can check your code, but you will have to refer to one of the resident image gurus w.r.t. features.

Greg Heath

Contact us