【作者】 郑建彬；

【导师】 朱光喜；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2006， 博士

【摘要】 签名作为人的一种行为特征,是表征个人身份的传统途径之一,具有很好的唯一性、非侵犯性、易为人所接受等特点,是一种公认的身份识别的技术。在线签名认证是通过计算机来采集和认证个人签名,从而实现自动身份认证的一种技术。设计制作了嵌入式在线签名采集系统。以单片机AT89S52为核心的在线签名数据采集系统,采用四线电阻式触摸屏传感签名信号,利用触摸屏控制芯片ADS7846可以采集笔迹的坐标及压力信息。设计了USB接口,通过USB控制芯片PDIUSBD12完成与PC机的通信。构造了在线签名数据库。组织了有规模的在线签名采集活动,搜集了40多人约5000个签名。真实签名数据库由真实签名构成。伪造签名数据库由随机伪造签名、熟练伪造签名和定时伪造签名构成。实现了一个实时在线签名认证系统,提出了两级认证机制。第一级认证采用基于参数特征的方法-基于签名能量特征的匹配。提出了一种以小波分析理论为基础的在线手写签名认证算法。基于Daubechies小波的方法对输入签名波形进行分解及部分重构,提取签名波形在跳变点的签名能量特征,抽取M个具有较大签名能量的跳变点。基于签名能量特征,提出了一种新的计算测试签名与参考签名相似性的算法。提出了直接按签名能量大小排序的比较法和基于DTW的方法对签名能量特征进行比较。本级认证主要是为了快速消除随机伪造签名。第二级认证采用基于函数特征的方法-基于签名曲线段的匹配。由于在线签名具有随意性,且将签名用函数表示十分复杂,在建立两个签名相似程度的比较准则的基础上,提出了签名认证的匹配模型,将匹配问题转化为函数优化问题。对于签名这种问题一般优化方法难以解决,由于演化计算只需要进行适应值的比较,可用来解决此类函数优化问题,提出了对测试签名与参考签名进行曲线段匹配演化算法。为了解决签名时存在时间轴的非线性问题,提出了签名曲线的动态分割匹配算法。针对演化算法产生新解无序的矛盾和算法设计中存在的搜索效果和效率平衡问题,引入了基于相似性的邻域搜索策略和利用适应值对个体进行分级的搜索策略。同时,为了提高搜索效率,对子种群进行加速以期找到较好的解集。本级认证主要是为了消除熟练伪造签名和定时伪造签名,提高签名认证的准确率。更多还原

【Abstract】 Signature, as one of the behavioral characteristics of human beings, is one of the traditional methods used to represent personal identity. Signature verification has become a well-known identity identification technology with its characteristics of uniqueness, dignity, and convenience. On-line handwriting signature verification is a new automatic personal identification technique through data acquisition and signature verification done by computer.An embedded on-line signature data acquisition system, whose core is the single chip microcomputer AT89S52, is designed. Signature signal is sensed through a 4-wire resistance touch panel, and signature coordinates and pressure information are collected through touch panel controller ADS7846 during the period of signature. A USB interface, whose control chip is PDIUSBD12, is designed for communication between the system and personal computer.An on-line signature database is constructed. Almost 5000 signatures from 40 subjects are collected in the well-organized on-line signature acquisition activities on a large scale. The genuine signature database is constituted totally by genuine signatures, while random forgery signatures, skilled forgery signatures, and timing forgery signatures constitute the forgery signature database.A real-time on-line signature verification system is accomplished, and a two-level verification mechanism is proposed.The first stage verification adopts parameter feature method, which is based on the matching of signature energy features. An on-line handwriting signature verification algorithm is proposed based on wavelet theory. First, by means of Daubechies wavelet decomposition of signature waveform, and after reconstruction of part signal, the energies of sharp trajectory change points in the signature waveform are extracted, and the M most dominant energies are chosen as feature vector. A new algorithm for evaluating the similarity between the testing signature and the reference signature is put forward, which is based on the signature energy feature. The comparison of energy features between the testing signature and the reference signature is made through the method that directly arranges in descending order the signature energies and the method that is based on DTW (Dynamic Time Warping). The presented algorithm based on energy feature is capable of eliminating quickly random forgeries for automatic signature verification.The second stage verification adopts function feature method, which is based on the matching of signature curve segments. Owing to the randomness of on-line signatures, and it is very difficult for functions to express the signature waveform. After the establishment of the similarity comparison criteria between the two signatures, the paper proposes the matching model of signature verification. As a result, matching problem can be converted into function optimal problem. A general optimal method is difficult to solve the kind of signature problem, but EC (Evolutionary Computation) is easy to do so for its comparison of fitness is only required. Based on EC, an algorithm of optimal curve segment matching between the testing signature and the reference signature is proposed. In order to solve the problem of time nonlinearity during the period of signature, the dynamic segmentation matching algorithm of the signature waveforms is proposed. As the new solutions generated by EC are out-of-order, and the balance problems between the search effect and efficiency exist in the algorithm design, the neighborhood search strategy based on the similarity and the search strategy based on the classification of the individuals by the fitness are adopted. In the meanwhile, filial-populations are processed through the acceleration operation in order to get better solution sets and to improve the search efficiency. In this stage, the main purpose is to eliminate the skilled forgery signatures and timing forgery signatures, and to improve the accuracy of the verification.更多还原