Off-line Signature Verification Incorporating the Prior Model


The existing signature verification systems usually train classifiers for a new user by both his/her genuine and forgery signatures. Obviously, the requirement of forgery signatures is impractical. This paper presents an off-line signature verification system that only requires the genuine signatures of a new user. At the training stage the system learns the mapping between the parameters of classifiers without simple forgeries and those with simple forgeries. In the application stage, a primary classifier is trained for a new user without his/her simple forgeries. The final classifier is obtained by transforming the primary classifier via the mapping learnt in the training stage. Experimental results confirm the effectiveness of the proposed system.

International Conference on Machine Learning and Cybernetics