A hybrid scheme for off-line Chinese signature verification

Based on characteristics of off-line Chinese signatures, a novel hybrid signature verification scheme with static and dynamic features extraction is proposed. Pseudo-Zernike invariant moments are used for static features due to their scale and translation invariance. When dynamic features are considered, firstly global and local HDFs (high-density factors) are obtained. Then weighted relative gravity center of global high density image is computed to generate another feature. In addition, an important feature is extracted using wavelet transform on normalized gray level histogram. Then ten orders of pseudo-Zernike moment invariants and seven dynamic features compose the eigenvector. At last for each class of signatures, a BP (back-propagation) network that was trained with 5 genuine signatures and forgeries that other signatures act as is built to verify unknown samples. A collection of 290 signatures is used to test the verification system. As a result of this experiment, the FAR (false acceptance rate) and FRR (false rejection rate) can achieve 7.84% and 6.89% respectively