【作者】 张大海；

【导师】 汪增福；

【作者基本信息】 中国科学技术大学 ， 模式识别与智能系统， 2008， 博士

【摘要】 随着社会的发展,越来越多的商业活动和工作实践被计算机化,个人身份的识别和认证在我们的社会生活中起着非常重要的作用。像电子银行这样的商业应用和门禁管制这样的安全应用场合都要求实时和准确地对个人身份进行识别。传统的基于密码和基于证件的个人身份识别和认证系统存在费时、使用麻烦、容易被遗忘、成本昂贵和容易被仿冒的弊端。生物特征识别技术,是一种基于人的解剖学特征(如人脸、指纹、虹膜等)和行为学特征(如在线/离线签名、语音、步态等)进行自动身份识别的技术。基于生物特征识别技术的认证可以克服传统的自动个人身份识别技术的局限性,目前在各个安全领域已经有了广泛的应用,但是,仍然需要研究新的算法和解决方案,以进一步提高性能。手写签名是日常生活中被广泛接受的一种个人身份认证方法,其优点主要有:签名的易于采集性,在任何使用签名的地方都可以使用;签名具有较好的稳定性和唯一性;签名的采集对人没有任何潜在的危害,为大众所接受;签名可以更换,用户可以像更改密码一样变更参考签名。签名认证按照签名的采集方式可以分为两类:在线签名认证和离线签名认证。其中在线签名认证可以综合利用签名的静态特征和较难模仿的动态特征,具有较好的认证效果。本文对在线手写签名的认证方法和应用进行了深入的研究,主要内容包括:(1)建立了基于F_Tablet的数据采集和处理平台,该F_Tablet设备不仅可以采集签名的字形序列(x,y),还可以获取签名的三维力信息序列(Fx,Fy,Fz)。基于该平台,采集并建立了签名数据库,并用该签名数据库开展了各种签名认证方法的实验研究。(2)从签名数据中提取了188个全局特征,这些特征中不仅包括多维力特征,也包括由此派生出来的字形特征。定义了特征重要性函数,对提取的特征进行重要性排序,选取那些有利于正确区分真伪签名的个性特征。对于选择出的个性特征,分别用基于支持向量机和基于隐马尔可夫模型的方法对签名进行验证,并将特征选择结果与用主分量分析和线性判别分析的测试结果进行比较,证明了该特征选择方法的有效性。此外,通过对大量签名的全局特征的观察,进一步提取了五个关键全局特征,利用这五个关键全局特征,可以实现非熟练伪签名的快速剔除。(3)提出了一种笔段分割方法。该方法通过综合检测签名的Fz方向压力的波谷点和字形的速度极小值点来实现。利用该方法将签名分段,并从每一笔段中提取签名的笔段特征。分别采用基于隐马尔可夫模型和基于串匹配的方法对签名的笔段信息进行了验证实验。实验结果表明,笔段特征可以有效地反映真伪签名之间的差别。(4)利用签名的力信息序列和字形信息序列,提出了一种基于改进的动态时间规整技术的签名验证方法,与普通的动态时间规整方法相比,错误率有明显的降低。将力信息序列和字形信息序列进行重采样后,利用基于主分量分析方法进行了签名验证实验。(5)综合签名的全局特征、笔段特征、力信息序列和字形信息序列,提出了基于多特征的在线签名验证方法,结果表明,多特征的融合可以有效地提高系统的性能,降低错误率。更多还原

【Abstract】 With the development of society,more and more business activities and work practices are computerized.Personal identification and verification play a critical role in our society.Ecommerce applications,such as e-banking,or security applications, such as building access,demand real-time and accurate personal identification. Traditional knowledge-based or token-based personal identification or verification systems are time-consuming,inconvenient to use,easy to be forgotten,expensive and easy to be imitated or copied.Biometrics refers to automatic recognition of people based on their distinctive anatomical(e.g.,face,fingerprint,iris,etc.) and behavioral (e.g.,online/off-line signature,voice,gait,etc.) characteristics.Biometrics based authentication can overcome some of the limitations of the traditional automatic personal identification technologies,but still,new algorithms and solutions are required to improve the performance.Compared with other personal verification methods,the handwritten signature is a well-accepted one in our daily life.Signature verification can be divided into two categories:online signature verification and offline signature verification.Since the former can make full use of both the static features and the dynamic features of a handwritten signature,it can obtain better verification results in practice.The work included in this dissertation focuses on the online signatures verification methods and their applications described as follows:(1) A platform for signature capturing and processing was built based on a F_Tablet which can not only capture the shape series(x,y) but also the three-dimensional force series(Fx,Fy,Fz),and a signature database was also constructed by using the F_Tablet.Meanwhile the studies on various signature verification algorithms were carried out.(2) 188 global features were extracted from the signature according to the multi-dimensional forces.These features included force features as well as shape features.The weight function of features was then defined and used to sort the features extracted and the personalized features that can help separate the genuine signatures from the fake ones were selected.Then the support vector machine based method and hidden Markov model based method were used to verify the features selected,and the corresponding results were compared with the results based on principal component analysis and linear discriminant analysis.The experimental results showed that the weight function of features had good effect.Moreover,five key global features were selected from further observation of all the global features of signatures.The experimental results showed that simple forgery could be easily and quickly detected by using the above five key global features.(3) The signature was divided into segments by the method that synthesized the valley point of the pressure and the local minimum point of the velocity.Segment features were extracted from every segment.We verified the signatures using methods based on hidden Markov model and based on string matching respectively.The results showed that the segment features could effectively represent the differences between the genuine and fake signatures.(4) An improved dynamic time warping algorithm was proposed to verify the force series and the shape series of the signatures.Compared to the usual dynamic time warping,the improved algorithm got better result.Resampled the force series and the shape series and then verified the signatures using the algorithm based on principal component analysis.(5) Finally,a multi-feature based online signature verification method was proposed.This method synthesized the global features,the segment features,the force series and the shape series.The multi-feature based decision fusion method proved to have better result.更多还原