Research Of On-line Signature Verification Based On Multi-stage DTW Matching

The technology of authenticate identity by biometric identification has been extensively researched and widely applied because it resolves the problem of "It can only recognize digit identity" in traditional authentication technologies. Handwritten signature verification is the most readily accepted way in all of the biometric identification technologies, thus it becomes a hot research field of pattern recognition. In on-line handwritten signature, the signature is presented by a series of timing characteristics, such as coordinates of each sampling point, pressure and direction angle. This paper will focus on the on-line handwritten signature verification technology.The procedure of on-line handwritten signature verification includes a series of process such as data acquisition, preprocessing, feature point detection, feature extraction and pattern matching. In this paper, each process in on-line handwritten signature verification is in-depth research, two new approaches as following is proposed in this paper:· A new method of normalize the rotary scales of signature is presented. In this method, the direction of the signature is calculated by using the center of the two parts of the signature curve, then rotate the direction vector to unified the directions of the two handwritten signatures.· A new method of feature point detection algorithm which is based on the movement of the center of gravity in neighborhood of sample points is proposed in this paper. By this method, stable feature points of the signature can be achieved. The problem of "It is hard to determine a common threshold for all signatures" is also solved by using this method.Dynamic Time Warping technology is a simple concept and robust pattern recognition algorithm which has been widely used to solve the problem of on-line handwritten signature verification. In this paper, a new method of on-line handwritten signature verification based on multi-stage DTW matching is proposed. Global feature, signature segment characteristics, sampling points characteristics are used in three stage of the signature verification to identify the signature from roughly to specific:· In the first matching stage, calculate the output probability of "global feature is matched between the test signature and the authentic signature", the output probability is compared with the preset threshold to exclude the significant pseudo signature.· In the second matching stage, the basic unit for feature extraction is signature segments, then calculate the signature segment DTW distance, the distance is compared with the prior training threshold to check whether the test signature is pseudo or not.· In the third matching stage, MMD algorithm which is proposed in this paper is used to obtained an equal number of signature segments between the test signature and the template signature; then calculate the DTW distance of each pair of corresponding signature segments, then calculate the distance between the whole signatures by distances between each pair of corresponding segments, the distance is compared with the prior training threshold to check whether the test signature is pseudo or not.