基于统计特征的维吾尔文离线手写签名鉴别技术研究

发布时间：2018-10-25 08:33

【摘要】：手写签名鉴别作为生物识别领域中的一种身份认证技术在金融、法律、商业等方面已被人们广泛地接受并应用。目前基于英文、阿拉伯文和中文的手写签名鉴别技术已经得到了较成熟的研究成果,而基于维吾尔文的手写签名验证在该领域内仍处于一个初始阶段。因此,对维吾尔文手写签名鉴别进行更深入地研究是弥补并完善我国少数民族离线签名鉴别技术体系中具有很大的实际应用和实用价值。本文主要是针对离线状态下的维吾尔文手写签名鉴别技术进行了研究和分析,其所完成的研究工作包括:签名样本的采集和预处理、特征提取、分类鉴别等三个部分。在预处理阶段通过灰度化、二值化、平滑去噪、归一化等方法来克服了签名图像上的噪声和干扰信号。在特征提取阶段,根据维吾尔文手写签名的书写风格及特点,对每一个签名样本图像分别进行四种不同的扫描来提出了一种16维的方向特征。其次,在方向特征的特征提取方法基础上统计出签名笔迹在6种不同方向上的签名笔迹黑像素点信息为基础提出了一种改进的48维方向特征。最后,以灰度共生矩阵的能量、熵、惯性矩、局部平稳性等特征参数为基础,采用特征加权融合的方法来提取了融合特征并确定了适合维吾尔文手写签名鉴别的最佳权值。在签名图像的分类鉴别阶段,对于本文提出的两种方向特征分别用欧氏距离、卡方距离和曼哈顿距离等三种距离分类器来进行了签名鉴别实验。而对于灰度共生矩阵加权融合特征利用BP神经网络来进行了签名鉴别。实验中,从维吾尔文手写签名样本库中选用了15个人(20个原签名样本/每人+20个简单模仿的伪签名样本/每人+20个熟练模仿的伪签名样本/每人)的900个手写签名样本。最后用本文应用的三种签名鉴别方法来得到的最高签名鉴别率分别为88.61%,96%和91.78%。
[Abstract]:Handwritten signature authentication, as a biometric authentication technology, has been widely accepted and applied in finance, law, commerce and so on. At present, handwritten signature authentication techniques based on English, Arabic and Chinese have obtained more mature research results, while Uighur based handwritten signature verification is still in an initial stage in this field. Therefore, the further study of Uygur handwritten signature authentication is of great practical application and practical value in making up for and perfecting the off-line signature authentication system of minority nationalities in China. This paper mainly focuses on Uygur handwritten signature authentication technology in off-line state. The research work includes three parts: signature sample collection and pretreatment, feature extraction, classification and authentication. In the preprocessing stage, the noise and interference signals on the signature image are overcome by grayscale, binarization, smooth denoising, normalization and so on. In the phase of feature extraction, according to the writing style and characteristics of Uygur handwritten signature, four different scans are performed on each signature sample image to propose a 16-dimensional directional feature. Secondly, based on the feature extraction method of directional features, an improved 48-dimensional directional feature is proposed based on the statistic of the black pixel information of signature handwriting in six different directions. Finally, based on the energy, entropy, moment of inertia and local stationarity of the gray level co-occurrence matrix, the feature weighted fusion method is used to extract the fusion features and determine the optimal weights suitable for Uygur handwritten signature authentication. In the phase of classification and authentication of signature images, three distance classifiers, Euclidean distance, chi-square distance and Manhattan distance, are used for the two directional features proposed in this paper. The weighted fusion feature of gray level co-occurrence matrix is verified by BP neural network. In the experiment, 900 handwritten signature samples of 15 people (20 original signature samples / 20 simple imitation pseudo-signature samples per person / 20 skilled imitation pseudo-signature samples / each) were selected from the Uygur handwritten signature sample database. Finally, the highest signature authentication rate obtained by the three signature authentication methods used in this paper is 88.61% and 91.78% respectively.
【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】