Off-line Signature Verification Based on Multitask Learning

Off-line signature verification is very important to biometric authentication. This paper presents an effective strategy to perform off-line signature verification based on multitask support vector machines. This strategy can get a significant resolution of classification between skilled forgeries and genuine signatures. Firstly modified direction feature is extracted from signature's boundary. Secondly we use Principal Component Analysis to reduce dimensions. We add sonic helpful assistant tasks which are chosen from other tasks to each people's task. Then we use multitask support vector machines to build a useful model. The proposed model is evaluated on GPDS and MCYT data sets. Our experiments demonstrated the effectiveness of the proposed strategy.