Asked by Monisha
on 3 Jul 2012

Please give me an idea on how to use this tool in Matlab! Any help would be appreciated!

Matlab Help is not helping much...

I need to have 250 images of every Kannada character as input and the target vector should consist of digits (indicating class) for each character. As in, 1 indicates it is the first character and so on...

The problem statement is basically recognizing handwritten Kannada characters. So any ideas on how to use neural networks for this and how to train it?

[Merged information from duplicate question]

How to create a dataset as input to a neural network for a character recognition system?

I have 250 images of each handwritten Kannada character that I need to feed into the neural network as input and my target vector should be such that the values within the target dataset should indicate which letter has been inputted. Maybe a 1 in a particular column can indicate the class? Any ideas? I have no clue. Please help.

Is this right? Suppose all 250 images of the first letter have a size of 75x75, does the input matrix have to be a size of 250x75x75? How do you do this? And if so, what should the target vector look like?

I am using nprtool. I have to create the target and input matrices as variables right? (with the extension .mat)

Please put my confusion to rest soon!

*No products are associated with this question.*

Opportunities for recent engineering grads.

## 3 Comments

## Walter Roberson (view profile)

Direct link to this comment:http://www.mathworks.com/matlabcentral/answers/42656#comment_87531

http://www.mathworks.com/matlabcentral/answers/?term=tag%3A%22ocr%22

## Monisha (view profile)

Direct link to this comment:http://www.mathworks.com/matlabcentral/answers/42656#comment_87659

Thanks :) but just a simple question how do I input 250 images of each character as input?

## Greg Heath (view profile)

Direct link to this comment:http://www.mathworks.com/matlabcentral/answers/42656#comment_92530

First you have to perform feature extraction to reduce the sizes from 75x75 = 5625 to something much smaller say, 10x10 = 100 (For realistic numbers search the internet for kannadi feature extraction). Then create an input matrix with subset matrices of size 100x250 for each character. These are concatenated to obtain a composite input matrix of size 100xN where N =250*c for c characters. The corresponding target matrix contains columns of the c-dimensional unit matrix eye(c). The row index of the "1" indicates the class index.

Hope this helps.

Greg