Using a Neural Network For Arbitrary Shape Recognition

    I used traditional character recognition techniques to write this program. First the input is translated into a binary 10 x 10 grid with a 1:1 aspect ratio. The neural network takes all the grid squares as input and has N outputs (number of known shapes.) Each output is the probability that the user inputted shape is that known shape.
    The program can also filter out noise from a scanned image. This is done by a convolution mask using either a mean or median filter.
    The neural network was trained using a back propagation algorithm. This demo was written in C# 2.0.