【作者】 刘蕾；

【导师】 段会川；

【作者基本信息】 山东师范大学 ， 计算机软件与理论， 2011， 硕士

【摘要】 手写签名鉴别是一种新兴的基于生物特征的身份识别方式,可以提供一种更为安全、可靠、便捷的身份识别的新途径,是人的生物特征中公认的、最容易被大多数人接受的一种身份认证方式,也是当前模式识别领域中的研究热点之一。手写签名鉴别首先是提取和分析签名信息的特征,然后根据相应的算法判定待鉴别签名与原有真实签名是否是同一个人书写的。其相关技术的研究对于电子商务和电子政务等领域的发展将起到巨大的推动作用,具有重要的理论意义和实用价值。动态时间规整作为一种手写签名鉴别方法,具有概念简单、算法鲁棒的优点,它将时间和距离测度结合起来,能对存在全局或局部扩展、压缩或变形的模式进行匹配,解决动态模式的相似度量和分类问题,并在参考模式R和测试模式T的签名特征信号之间找到一条优化的时间校准匹配路径。但这种方法假定每一点都同等重要,在签名鉴别中误差很大。由于在签名轨迹上各点的重要性不同,本文提出给每个采样点加入权值,加强稳定点(重要采样点)对DTW的贡献,降低非稳定点(次重要及不重要采样点)对DTW的影响。主要是采用将加速度、压力、加速度和压力的组合作为权值特征,通过比较识别率得到一个最优的权值特征。实验数据表明,压力在手写签名的鉴别中被认为是最优的权值特征。方差可以反映数据相对均值散步的程度,我们在匹配距离的基础上运用方差公式计算不同签名的类内距离、类间距离。然后通过比较类内距离、类间距离的大小就可以得出不同类在书写时的差别,据此就能够更大程度的提高签名鉴别率。本文对手写签名鉴别的各主要阶段都开展了一定程度的研究,这些阶段包括:数据获取、数据预处理、特征提取、签名匹配。数据获取阶段主要采用WACOM手写板采集相关的实验数据。由于签名者的签名具有随意性,即可以在手写板的任意位置进行签名,而且在签名获取的过程中会伴随着一些噪声产生,因此为了提高识别率,在签名鉴别前首先进行预处理。预处理阶段所采用的主要方法有去除噪声点、将坐标原点归一化到质量中心、平滑等。在识别对象时,特征是唯一的依据,特征选取时通常遵循的原则是使该特征具有代表性和方便计算。算法的选取对最终的结果也有很大的影响,所以在签名匹配这一步我们详细介绍了加权DTW较传统DTW的优势,并通过比较实验结果得出结论:加权DTW较一般的DTW在识别率上有明显的提高,权值特征的选取对签名的鉴别有着十分重要的意义。更多还原

【Abstract】 Handwritten signature verification is an emerging identification method based on biometrics. It can provide a more secure, reliable, and convenient new ways of identification and it is the most easily identity authentication recognized in public in a person’s biological characteristics. It is also one of the current hot researches in the field of pattern recognition. Handwritten signature verification first extracts and analyzes the characteristics of signature information. Then determines the signature to be signed with the real signature is written by the same person based on the corresponding algorithm. The related technology research plays a significant role in promoting e-commerce and e-government developments and has important theoretical significance and practical value.As a handwritten signature verification method, Dynamic Time Warping has advantages of simple concept and robust algorithm. It combines time and distance measure to match global or local, compression or deformation patterns to solve the similarity measure and classification problems about dynamic model, and finds an optimal calibration path between reference mode signals R and test mode signals T. However, this method assumes that each point is equally important, so great error appears in signature verification.For these reason, we will add weight to each sample point, which will strengthen the contribution of stable points (important sampling points) to DTW, Simultaneously reducing the effects of unstable points (sub-sampling points). We mainly use acceleration, pressure and the combination of the first two as weights, by comparing the recognition rate to find an optimal feature. Experimental data indicated that pressure is the best feature in signature verification.Variance reflects the extent of the data relative to the mean. We first calculate the intra-class distance and intra-class distance for different signature using the variance formula based on the matching distance, then get the variance for different classes in writing by comparing the intra-class distance and inter-class distance, hereby we can get greater rate of increase in signature verification.In this paper, we carried out research to some extent about all the major stages of the handwritten signature verification. These stages include: Data Acquisition, Data Preprocessing, Feature Extraction and Signature Matching. In the step of Data Preprocessing, we used WACOM tablet to collect experimental data. Since the signature of the signer is arbitrary, that is, one can signature anywhere on the tablet. Meanwhile, a number of noises generated in this process. In order to improve the recognition rate we should preprocess the sample data before we verify the signature. The main method is removing noise points, putting the coordinate origin to the center of mass normalization, smoothing and so on. The feature is the only basis in the identification of the object. We should select representative and convenient feature information in our calculation. Algorithm selection has a great impact to the final result, so in the step of signature matching we introduce the advantages of weighted DTW compared with traditional DTW in detail, and conclude by comparing the experimental results: the weighted DTW has apparently improved recognition rate, and the introduction of weight features is of great significance in handwritten signature verification system.更多还原