Performance Analysis of Different Types of Neural Network for Signature Verification System

Signature may be used as biometric and is also suppose to be distinctive for every individual just like thumb impression, iris etc. There are so many methods available for signature verification, online as well as Offline. Use of Artificial Neural Network (ANN) is one of them. In ANN also many types of neural network may be designed for the signature verification system but which type suits the best is still a question. In this paper author analyzes and suggests the option in neural network which best suits for offline signature verification system. For this four options i.e. trainable cascade-forward backpropagation network, Elman backpropagation network, feed-forward backpropagation network, feed-forward input-delay backpropagation network are taken into consideration and a performance analysis is done using chain code method of offline signature verification system. The performance is compared on the basis of time required for training, accuracy, False Acceptance Ratio and False Rejection Ratio. It is observed that for less number of training samples (up to 30 samples) the Elman backpropagation method seems better and for more than 30 training samples the feed-forward backpropagation method is best.