On-line Handwritten Signature Verification Based on Two Levels Back Propagation Neural Network
Zhan En-qi, Guo Jin-xu, Zheng Jianbin, Ma Chan, Wang Lin-juan · 2009
A new algorithm for on-line handwriting signature verification is proposed. 15 statistic features is used for representing the signature, which extracted from the handwritten signature, at the same time, applied as the input of Back Propagation Neural Network. Via the Daubechies-6 wavelet transforming, 32 wavelet decomposition coefficients of X coordinate and Y coordinate are selected as wavelet features. The two levels verification method is considered. The first level adopts statistic features, and wavelet features as the second. Through verifying five groups signature, FRR (False Rejection Rate) of the whole samples verification arrives at 6%, and FAR (False Acceptance Rate) is 1.07%. The experimental results show the efficiency and feasibility of this algorithm.