Recognizing digits from neural networks

I am doing a project to read digits from a calculator screen. To recognize the digits, I intend to use neural networks. I have to train the network first with about 10 or more sample images of all digits(0 to 9) and then go on to identify them. Can anybody suggest how to input the training images into the neural network and in what format? I know the basic concept of neural networks but cannot implement it in a situation like this.

 Accepted Answer

Greg Heath
Greg Heath on 3 Apr 2013
input =image(:); % columnize
Thanks for formally accepting my answer

7 Comments

Greg, would you please elaborate a little more? Do I have to columnize the samples of 1 digit in a matrix and create 10 such matrixes for 10 different digits or the 10*10 samples of all the 10 digits in one single matrix? I am kind of a rookie in matlab actually.
Each training image should be reshaped into vector form, to create a single "sample" for NN purposes. The class you would use for the "sample" would be the associated digit value.
Okay, I have to make 10 such classes or matrices for 0-9. Suppose the 10 training images for digit 0 are im00, im01, ..., im09, similarly for digit 1 are im10, im11, ... im19 and likewise. How can I columnize these images and train them and later on recognize them from new images? Could you provide me some sample code?
Can you provide me some code how to train the networks and how to save it? Any web resources regarding the same would also be useful. Thanks.
and what to specify as the target matrix? The pattern recognition tool asks for a target matrix.
I have a similar kind of problem for recognition purpose. Would please share your neural network code? @Soumyadip

Sign in to comment.

More Answers (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!