Offline handwritten signature verification using OC-SVM and BC-SVM classifier

Handwritten signature remains to be the easiest form of identity authentication used in modern living; including banking, legal, financial transactions and others. Thus, a robust and efficient handwritten signature verification system still plays a key role in data security. This study presents the use of SVM, both one-class (OC-SVm) and bi-class (BC-SVM) to be used in signature verification. The experiment was conducted with random respondents who were asked to write their identical genuine signatures and forge signatures of another respondent. The collected images were then subjected to image processing to fine tune the features of the signature before subjecting the image to feature extraction. Feature extraction was carried out by zoning, rather than as a whole, to determine the features precisely. There were two verification system models trained and tested using the regional and geometrical feature vectors, the BC-SVM and OC-SVM model. The study resulted with high accuracy performance in verifying the signatures with low False Acceptance Rate (FAR) and False Recognition Rate percentage (FRR).