Online Signature Recognition using Neural Network

Abstract In this work a new method of signature recognition with neural network is proposed. The features are extracted from two raw databases of ATVS signature database: one consisting of 25 signature samples of 350 persons and other 46 signatures of 25 persons. The features include 9 features computed by D.S. Guru and H.N. Prakash and proposed features are no of pen downs, magnitude of average velocity, magnitude of average acceleration and length to height ratio. Signature features are pre-processed with a scaling method and brought to a value having same decimal point. A feed forward neural network is trained using back propagation learning method. With each features removed, an accuracy rate is calculated to check the feature which will be better for signature verification. Accuracy of recognition up to 98% and 89% are obtained using signature samples of 10 persons from each database respectively.