A Study of On-line Signature Features Selection and Verification

• Main Authors: Li, Pei-Ru, 李珮如
• Other Authors: Tsou, Chi-Ming
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/75669385255503298186

碩士 === 龍華科技大學 === 資訊管理系碩士班 === 103 === How to verify and confirm the signature of an electronic document is an important study issue for the on-line e-commerce transactions. Most of the studies for signature verification are conducted by collecting the signature data, usually 5 to 10 copies from tester then try to find out and verify the major factors which cause the variation of signature from the signature data set. However, only single copy of on-line electronic signature can be collected in e-commence environment, therefore, how to enhance the precision of the on-line signature verification under the condition of collecting only single copy of electronic signature is the main purpose of this study. A signature capture program which built on a smart phone is designed to collect the signature data that are conducted by a group of testers and incorporate a signature verification procedure to examine the signature features. Each participator was requested to provide 10 genuine signatures and incorporate others signatures which are extracted randomly to build the training datasets. During conducting the signature, the program will collect the position and time information; then a preprocessing procedure is applied to extract static and dynamic features from the signature data afterwards, and then incorporates global features retrieving and local pattern similarity algorithm those two methods to conduct on-line signature verification by collecting normal and simple or mimic forgery signature with smart phone. The collected signature data sets are divided into training and testing data sets randomly after pre-processing for experiment. The training data sets are used for building the model with Generic algorithm (GA), Support Vector Machine (SVM) and Artificial neural network (ANN) those three data analysis methods. The testing data sets are then verified against the model for verification afterwards for simple and mimic forgery signature respectively. The experimental results show that for the simple forgery signature, the performance of SVM is the best among the three methods; the value of the average deviation is 0.016. Moreover, for mimic forgery signature, the performance of ANN is the best among the three methods; the value of the average deviation is 0.0475.