서명검증에서 특징값을 고려한 판단 경계 설정에 관한 연구

Determination of Decision Boundary Using Feature Values in the Signature Verification

Usually, more reference signatures result in better performance in signature verification. However, registering .many signatures may be a tedious work for users, so algorithms that use less signatures for the registration without increasing error rate is needed. In this paper, we find the features such as pen-down duration, the number of locally minimum velocity points, and the number of locally maximum curvature points. Then we find the relationship between these features and the optimal decision boundary. We apply this relationship in deciding threshold for signature verification. Experimental results show that the method using three reference signatures has almost same error rate as algorithms with many references.