A STUDY OF THE STABILITY OF SIGNATURE FEATURES BY NONLINEAR LOCAL OPTIMAL TIME WARPING

Using dynamic information of signing to verify signatures can raise the safety of verifying systems. It is a problem of vector classification in the space spanned by the dynamic features of handwritten signatures. However, the discrete-time signals of sample signatures are usually warped to different extent at different positions, which distorts the distribution of the feature vectors obtained from these signals in feature space. Therefore, it is difficult to determine the stable subspace that separates the genuine signatures from forgery ones. This situation becomes worse when the total number of sample signatures is low. In this paper, a nonlinear local optimal time warping algorithm is put forward, which has low computational complexity but with good performance. It can effectively increase the compactness of the feature vectors of genuine signatures, and therefore enlarge the distance from the feature vectors of forgery ones. Through using the nonlinear local optimal time warping, together with the linear time warping, the feature space can be divided into several subspaces. Each subspace indicates genuine signatures with different confidence level. This enables the signature verification system to satisfy the requirement of different security.