query

[manoj]

I am working on classification of normal and cancerous brain mri
images.I extracted statistical features from the images and i got some
features from the cooccurence matrix too.Totally there are ten
features.But i had got only 12 images downloaded from net.6 normal
and 6 cancerous.but when i gave these ten features to a feed forward
network and tried for classification the accuracy found to be very
less.i thought the accuracy may be less because of less number of
training samples.so i used same samples again and again and now i made
12 images to 96 images by repeatedly giving the twelve images 9
times.now the classification effeciency has improved very much and the
network takes more number of test samples when compared to the first
case.but i dunno whether it is correct to train the sample repeatedly
with the same samples.I want some one to guide on this aspect.I have
used functions from matlab 7.5 where the function itself divides the
input samples into training and testing.Can anyone suggest me how to
give training and testing samples seperately for a network?
thanks in advance
manoj