Thanks for your response,
my network needs to classify facial images into one of 4 expressions - neutral, happy, sad and suprised.
I have implemented a feed forward backprop network but results not great, unsurprising as I have only been using default settings so far.
I was thinking that I may put my images through a self organising map and then input the results into my ffbp network but dont know how useful this will be.
Cheers