基于LSVM分类鉴定器的脱机签名鉴定研究  

ON OFF-LINE SIGNATURE VERIFICATION BASED ON LSVM CLASSIFIER

作　　者：朱浩悦[1] 耿国华[2] 周明全[2] 李佳[3] 

机构地区：[1]西安文理学院计算机科学与技术系,陕西西安710065 [2]西北大学可视化研究所,陕西西安710069 [3]中国电子科技集团公司第二十研究所,陕西西安710069

出　　处：《计算机应用与软件》2009年第7期219-221,共3页Computer Applications and Software

基　　金：西安文理学院专项科研基金资助项目(KYC200738)

摘　　要：针对脱机中文签名鉴定,主要对脱机签名鉴定的特征抽取和比较决策做进一步的研究。在特征提取与选择上,在参考国内外一些成熟方法的基础上做相应的改进和尝试,使用静态形状特征和伪动态特征相结合的方法,提出一种新的高灰度稳定区特征,在特征选择上采用一种把概率距离法中的Bhattacharyya距离和特征本身综合起来考虑的方法;在比较决策上,采用比标准SVM算法速度更快,更易于实现的LSVM算法作为分类鉴定的方法,取得了较好的效果。This paper is about off-line Chinese signature verification. In this paper, its feature extraction and comparison decision-making is mainly discussed. In feature extraction and selection, some corresponding improvements and attempts are made based on some ripe methods developed at home and aboard. The signature＇ s static shape feature and pseudo dynamic feature have been Unified in feature extraction, a new high gray stable area feature has been proposed. A comprehensive consideration method which combining Bhattacharyya distance in probability distance method and the feature itself together has been used on feature selection, and the Lagragian Support Vector Machine classifier is taken as the classification and verification method on comparison decision-making, which runs faster than standard support vector machine and is easier to implement. Using this method, we have obtained better results.

关 键 词：脱机中文签名鉴定 生物测定 特征提取 拉普拉斯算子 LSVM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S433[自动化与计算机技术—计算机科学与技术]