脱机手写签名鉴别研究  

STUDY ON OFF-LINE HANDWRITTEN SIGNATURE VERIFICATION

作　　者：左文明[1] 

机构地区：[1]华南理工大学电子商务学院,广东广州510640

出　　处：《计算机应用与软件》2005年第9期102-104,共3页Computer Applications and Software

摘　　要：本文主要研究了脱机手写签名的特征提取,提出了一种结合静态特征与动态特征的新的鉴别方法。提取静态特征时,利用伪Zernike矩的尺度及位移不变性,在细化的签名图像上计算10阶伪Zernike不变矩来组成特征向量。提取动态特征时,则首先从灰度图像得到签名的全局及局部高密区域,利用高密区域与原签名图像对应部分的面积之比得到全局和局部HDF。另外在全局高密区域的基础上,计算其相对重心,并将其作为另一个特征。结合两类特征形成16维特征向量后,建立一个系统,在系统中采用290个真伪签名进行验证。实验结果表明,系统的FAR和FRR分别可以达到7·25%、9·30%。In this paper, features extraction of off-line handwritten signature is mainly discussed, and a new method with static and dynamic features extraction for verification is proposed. Scale and translation invariance of pseudo-Zernike moments is used for static features extraction. 10 orders pseudo-Zernike moment invariants computed based on thinned signature image are used to compose eigenvector. When dynamic features are extracted, first global and local HDRs are obtained from gray level image, then global HDF and local HDF are computed as ratio of HDRs area to the corresponding signature image area. In addition, based on global high density image, relative gravity center is calculated as another feature. 290 signatures are used for verification and experiments result shows that FAR and FRR can be up to 7.25% and 9.30% respectively.

关 键 词：签名 脱机签名鉴别 伪Zemike矩 高密区域 签名鉴别 手写签名 脱机 ZEMIKE矩 特征提取 动态特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP391.41[自动化与计算机技术—计算机科学与技术]