how characters recognition works using opnencv ANN_MLP ?

edit

i am so confused on how data is pushed to the training part for neural network to do it's job. so lets say i have 10 digits and 20 characters and for the training part each has 3 images so tottal of 30+60=90 images my question is

1- how is the images supposed to be stored in my files for instance are all the 90 images stored in one file and for the training a for loop will iterate on them ? 2- are all the images have to be the same size. also the recognized image?(input image)

side note: i have done SVM before but this is different since i have more than one image to be recognized SVM just say if its a dog or not but neural network says if its a dog ,cat or cow

any links on how implementation using java and opencv would be helpful