Application of a Steerable Wavelet Transform using Neural Network for Signature Verification

This paper describes a technique for the problem of off-line signature verification using a steerable wavelet transform. The signature has been treated as a two-dimensional image and uses the wavelet as a tool of data reduction and feature selection. Feed-forward neural network based architecture is used for both training and classification, because of the generalisation, fault tolerance, and such other capabilities of the neural network. Besides, the small length of the wavelet coefficients vector, which is used as a feature vector reduces the complexity of the neural network in terms of the number of neurons and time of training. Experimental results based on 300 signatures from 30 persons are presented, which show that wavelet has great potential for off-line signature verification.