基于特征点统计模型的联机签名校验

Online signature verification based on statistical modeling of characteristic points

在联机签名校验中,动态时间规正(DTW)方法是一种常用的校验算法,在非线性时间对齐的基础上给出两个签名间的距离并进行判决,这样做经验的成份较多,缺乏统计基础。该文提出了签名的特征点统计模型,利用DTW算法在序列匹配的基础上从签名中提取到多个特征点,将每个特征点的变化情况描述为多维统计特征的概率分布,在所有特征点具有同样协方差分布的假定下得到具体的概率分布参数。按照此模型推导出了在最小风险准则下对签名进行真伪判决的判决准则。采用此方法对一个公共的签名样本库进行了真伪校验测试,得到了4.41%的等误率。

Dynamic time warping （DTW） is a commonly used method for online signature verification. Dynamic time warping makes a decision based on nonlinear aligning of the two signatures which produces a DTW distance, but the method is subjective and lacks a statistical basis. This paper presents a method to model a signature＇s characteristic points. The DTW method is then utilized to match the two signature sequences so that the characteristic points can be extracted from the matching result. Changes in a characteristic point are described by a multi-variable statistical probability distribution. The probability distribution parameters are calculated assuming that all characteristic points in a signature have the same distribution. A statistical discriminant function is used to judge whether a signature is genuine or forged based on the minimum potential risk. The proposed method was tested with a general signature database with an error rate of 4.41%.