Neural network inputs and targets

[roldy]

I've written a program for character recognition but now upon testing I am unsure about a couple things. I'm having problems with training and simulating the network. Here is my interpretation of how network training goes.

Suppose I want to recognize the letters in the ABCD image attached. In MATLAB I used the function

[net,tr] = train(net,inputs,targets);

To train I set inputs = targets, where targets is the image with "ABCD". Then to recognize, say the letter D, I use this function

sim(net,input);

where input = the letter "D".

All this is done after appropriate image processing where I convert the images to binary matrices where each row describes a letter.

Is my understanding in how MATLAB trains correct? I want to be able to train once for all the letters, then have the user select any image (each image contains either only a "A", "B", "C", "D") and submit it to the network and notify the user if there is a match.

 

[Future Bruno]

Yes, your understanding is correct. The train function is used to create a neural network model by fitting a set of training data (inputs and targets). Once the network is trained, you can use the sim function to test the network with new data (inputs) and see how it performs. To recognize different letters, you need to create separate input matrices for each letter. Then, when you submit an image to the network, you can compare the output of the sim function with the inputs you used to train the network and determine which letter is present in the image.