基于签名视频的身份识别方法研究

Bank card (credit or debit card) based transaction has become the main means of payment in modern society. The security of card based transaction directly determines the capital and property security in the transactions. Card fraud (i.e.transactions with the stolen cards) is the main security threat of the card based transactions. Rapid and accurate personal recognition is the effective means to prevent the card fraud. Because the card based transactions require the user to confirm the transactions by his signature, it is the most natural and the most widely accepted way to authenticate a person by using his signature in the transactions. However, the current signature recognition cannot be widely used in the card based transactions due to its defects. Therefore it is difficult to prevent card fraud, resulting in huge economic losses. In order to solve this problem, this project will acquire and analyze the signature video, and recognize a person by using a variety of biometric features in the video, such as the signature behavior, the pen holding posture, the signature image and the texture of hand, etc. The research contents include the extension of the current signature video database, segmentation or extraction of the interest of targets in the video, feature extraction and learning, feature matching and recognition, etc. The research of this project is expected to establish a complete theoretical framework for personal recognition based on signature video, improve recognition accuracy and user acceptance, so as to promote its application in card based transactions and effectively prevent the card fraud.

刷卡交易已成为现代社会最主要的支付手段。刷卡交易的安全直接决定了交易中资金和财产的安全。刷卡欺诈（即盗刷别人的银行卡）是刷卡交易面临的最主要的安全威胁。快速准确识别刷卡人的身份是有效防止刷卡欺诈的有效手段。由于刷卡交易需要用户签字确认，利用签名来认证用户的身份是最为自然、也最被广泛接受的方案。然而，目前的签名识别由于自身的缺陷而未能在刷卡交易中得到广泛应用，很难防止刷卡欺诈，从而造成巨大的经济损失。为了解决这个问题，本课题将获取和分析签名视频，综合使用视频中的多种人体信息，如签名动作、握笔姿势、签名图像、手部纹理等，来完成高精度的实时身份识别。研究内容包括签名视频数据库的扩充、感兴趣目标的分割或提取、特征提取和学习、特征匹配与识别等。通过本课题的研究，旨在为基于签名视频的身份识别建立完整的理论框架，提高签名识别的精度和用户接受度，从而推进其在刷卡交易中的应用，有效预防刷卡欺诈的发生。

刷卡交易已成为现代社会最主要的支付手段。刷卡交易的安全直接决定了交易中资金和财产的安全。刷卡欺诈（即盗刷别人的银行卡）是刷卡交易面临的最主要的安全威胁。快速准确识别刷卡人的身份是有效防止刷卡欺诈的有效手段。由于刷卡交易需要用户签字确认，利用签名来认证用户的身份是最为自然、也最被广泛接受的方案。然而，目前的签名识别由于自身的缺陷而未能在刷卡交易中得到广泛应用，很难防止刷卡欺诈，从而造成巨大的经济损失。为了解决这个问题，本课题将获取和分析签名视频，综合使用视频中的多种人体信息，如签名动作、握笔姿势、签名图像、手部纹理等，来完成高精度的实时身份识别。本项目的主要工作包括：（1）签名视频数据库建立方面：扩充签名视频数据库至20000 段视频，使之包括300 个不同用户在不同时间的签名视频；（2）签名视频感兴趣目标分割理论与算法方面：提出了多种感兴趣目标分割算法；（3）签名视频特征提取和学习方面：提出了多种特征提取和学习方法；（4）签名视频身份识别方面：基于不同的特征，提出了多种分类算法，能够有效地进行身份识别。主要成果有：建立了一套包含20000段视频的专业数据库，实现了一基于签名视频的签名识别原型系统，发表/录用17篇SCI/EI检索的高水平论文，申请/获得专利和软件著作权共9项，培养研究生16人。本项目圆满完成申请书所列的研究内容，取得的成果在相关的图像处理、模式识别理论和技术方面取得了突破，同时促进了签名视频认证技术的发展和成熟，具有重要的理论意义与应用价值，使得实时在线签名认证大规模普及成为可能，可以节省大量的人力以及资源的浪费，从而减少信用卡盗刷的风险。