How to decide inputs and targets for neural networks for a signature recognition and verification system?

Hello!

I am doing project on offline signataure verification using neural network. I have prepared the database of 360 signatures(8 genuine and 4 forge signatures of each of the 30 person) and extracted features (moments of image using Zernike moments) of each signature. But I dont know how to train the neural network so that it can recognize the genuine and forge signatures. thanks.

Answer 1

[Greg Heath]

For N=360 examples of I-dimensional extracted feature column vectors and corresponding N-dimensional row vector of class indices i (1<=i<=c=30), the target matrix is ind2vec(indices) and

 [ I N ] = size(input)  % [ I  360 ]
 [ c N ] = size(target) % [ 30 360 ]

The default number of training vectors is

 Ntrn = N -  2*round(0.15*N) % 252

yielding

 Ntrneq = Ntrn*c     % 7560

training equations. When the number of hidden nodes, H satisfies

 H << Hub = -1+ceil( (Ntrneq-c)/(I+c+1))

The the number of weights

 Nw = (I+1)*H+(H+1)*c

is much less than the number training equations. Otherwise validation stopping and/or regularization are recommended.

Hope this helps.

Thank you for formally accepting my answer

Greg