基于稀疏表示的签名真伪鉴别方法研究

发布时间：2018-02-13 14:03

本文关键词： 离线签名鉴别改进高斯滤波改进Ostu CNN 稀疏表示　出处：《西安科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着现代化信息技术突飞猛进的发展,人们逐渐适应并习惯了由各种信息网络组成的全面数字信息化的生活方式。在此大背景下,身份认证已成为人们的日常生活中不可或缺的一部分。目前签名是很多领域中实际使用率最高的一种生物特征,人们对将签名用于虚拟经济的移动支付上的渴望也逐渐加深,近几年来对签名鉴别系统的研究也已成为热点。本文主要研究的是离线手写签名的分类鉴别,主要的研究过程可以分为四个阶段:签名图像样本采集阶段、签名图像样本预处理阶段、签名图像样本特征提取阶段和不同方法分类鉴别阶段。本文主要的创新点如下:(1)为更好的对签名图像进行平滑滤波,克服传统高斯滤波的不足,更好的表现其尺度不变性和旋转对称性,本文基于传统高斯滤波提出了一种改进的高斯滤波方法,实验结果表明,改进高斯滤波进行平滑去噪的效果比其他三种方法好,说明提出的改进方法是可行的。(2)为更好的对签名图像进行二值分割,自适应地找到更合适的阈值,在Ostu的基础上提出了改进。通过签名图像二值化实验,结果证明改进Ostu二值化的效果比其他三种方法好,说明提出的改进方法是可行的。(3)为了能更好的用最少的、最合适的特征去表示一个签名的全部特征,本文利用CNN结构在训练参数的过程中会自己根据数据学习并表示签名特征的特点,提出了一种基于CNN特征提取和稀疏表示的融合算法进行迭代和分类决策,实验结果表明此融合算法得到的测试集分类准确率平均提高到了 98.2054%,是四种算法中分类准确率最高的,说明此算法是可行的。系统实验结果表明,本文设计并实现的离线手写签名分类鉴别系统是稳定可靠的,为在线签名分类鉴别系统的研究和在移动设备的应用上提供了一个良好的方向。
[Abstract]:With the rapid development of modern information technology, people gradually adapt to and get used to the comprehensive digital information life style composed of all kinds of information networks. Identity authentication has become an indispensable part of people's daily life. At present, signature is one of the most widely used biometric features in many fields, and people's desire to use it for mobile payment in virtual economy is becoming more and more serious. In recent years, the research on signature authentication system has also become a hot topic. This paper mainly studies the classification and authentication of off-line handwritten signatures. The main research process can be divided into four stages: the phase of signature image sample collection. Signature image sample preprocessing stage, signature image sample feature extraction stage and different method classification stage. The main innovation of this paper is as follows: 1) in order to better smooth filter signature image, overcome the shortcomings of traditional Gao Si filter. The scale invariance and rotation symmetry are better represented. Based on the traditional Gao Si filter, this paper puts forward an improved Gao Si filtering method. The experimental results show that the improved Gao Si filter is more effective than the other three methods in smoothing noise removal. It is shown that the improved method is feasible. In order to better binary segmentation of signature image and find more suitable threshold adaptively, an improvement is proposed on the basis of Ostu. The results show that the effect of improved Ostu binarization is better than the other three methods, which shows that the proposed improved method is feasible.) in order to better use the least and most appropriate features to represent all the features of a signature, In this paper, a fusion algorithm based on CNN feature extraction and sparse representation is proposed to make iterative and classification decision based on the characteristics of data learning and signature feature representation in the process of training parameters by CNN structure. The experimental results show that the average classification accuracy of the test set obtained by the fusion algorithm is 98.2054, which is the highest among the four algorithms, which shows that the algorithm is feasible. The off-line handwritten signature classification and authentication system designed and implemented in this paper is stable and reliable, which provides a good direction for the research of online signature classification authentication system and its application in mobile devices.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】