CASIA OpenIR  > 毕业生  > 硕士学位论文
脱机手写签名验证方法研究
幸子健1,2
Subtype工学硕士
Thesis Advisor刘成林
2018-05-23
Degree Grantor中国科学院研究生院
Place of Conferral北京
Keyword脱机手写签名验证 Siamese网络 特征提取 度量学习
Other Abstract

手写签名作为一种重要的身份认证手段已经被用于许多行业,如各类文书、合同、银行票据等。目前脱机签名验证的方法,主要分为两类,即书写者相关和书写者无关。考虑到实际应用部署的条件,本文基于书写者无关的策略,对脱机手写签名验证的问题进行了研究。本文使用Siamese卷积神经网络解决脱机签名验证的问题。与现有的方法将特征提取和度量学习作为两个独立阶段的策略不同,我们采用深度学习的框架,将两个模块统一到一起,并可以进行端到端训练。在两个脱机签名验证公开数据集(GPDSsynthetic和CEDAR)上的实验结果证明了我们的方法在脱机手写签名验证问题上的优越性。本文主要工作和结果如下:
(1)对传统的基于特征提取和度量学习的签名验证方法进行了实验评价。本文详细介绍了轮廓波变换(Contourlet transform),Local Binary Pattern(LBP)特征,以及Histogram of Orientation Gradient(HOG)特征,并用支持向量机做书写者无关度量,在CEDAR和GPDSsynthetic这两个数据集进行实验,在大型数据集GPDSsynthetic上的结果可以作为传统特征的benchmark。
(2)基于Siamese网络的方法,将特征提取和度量学习结合到一起,参考近年来经典的网络结构的设计理念,重新设计了网络结构,在GPDSsynthetic和CEDAR这两个数据集上均取得了不错的性能。

 

;

Handwritten signatures have been used popularly in daily life as an important means of identity authentication, for various types of documents, contracts, and bank checks. There are two main approaches for signature verification: writer-dependent and writer-independent. Considering the requirements of practical applications, we study into methods for writer-independent offline handwritten signature verification. We present an offline signature verification method using Siamese convolution neural network. Unlike the existing methods which design feature extraction and metric learning in two independent stages, we adopt a deep-leaning based framework which combines the two stages together and can be trained end-to-end. Our experimental results on two public datasets (GPDSsynthetic and CEDAR) demonstrate the superiority of our method on the offline handwritten signature verification problem. The main work and results of this thesis are as follows:
(1)
We evaluate some representative traditional signature verification methods based on feature extraction and metric learning. This thesis introduces the Contourlet transform, Local Binary Pattern(LBP) features, and Histogram of Orientation Gradient(HOG) features in detail, and uses the support vector machine as a writer-independent metric method to perform experiments on two public datasets, CEDAR and GPDSsynthetic. The results on the largest dataset GPDSsynthetic can be used as benchmarks for traditional features.
(2)
Based on the Siamese network , we combine feature extraction and metric learning in a unified framework. We design neural network architecture based on some classic network structures. Our method achieve good performance on the GPDSsynthetic and CEDAR datasets.

Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/21187
Collection毕业生_硕士学位论文
Affiliation1.中国科学院自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
幸子健. 脱机手写签名验证方法研究[D]. 北京. 中国科学院研究生院,2018.
Files in This Item:
File Name/Size DocType Version Access License
Thesis_openir.pdf(6539KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[幸子健]'s Articles
Baidu academic
Similar articles in Baidu academic
[幸子健]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[幸子健]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.