binarization of an image is n^2.
blob detection is possibly n^2.
computation of image moments is also n^2.
feature extraction is also n^2.
clssification takes a long time given that our backpropagation
has 32-100-4 input, hidden and output units respectivelty.
is it even possible for us to solve a real-time character recognition
problem in less than 10 frames per second?