Classifying 2D patterns (images of alphabet letters) using Perceptron rule

hi i have asked this question couple of times in different forms but didnt get any real answers

i have 520 images of alphabet letters (20x26) and need to use single layer perceptron to classify them using LOOCV (leave one out cross validation)

my images are grayscale but i need to turn them into bipolar values (-1 and 1) this part isnt a problem

i know i need to use newp to creat the net and train it

but i have no idea how or where to even start

id like to at least know an algoritem or guide line of where to start and where to get

im not a computer student im EE and no idea how or even what i need to do

thank you