Online Handwritten Verification Algorithms Based on DTW and SVM

Abstract

With the development of computer network and biometric technology, identity verification with biologic characteristic becomes the need in this information era. Hand-written signature verification is a kind of the most acceptable way of identity verification, and as one kind of the most common method of authorization with a long history. Based on DTW (Dynamic Time Warping) and SVM (Support Vector Machine), we present a new algorithm for online hand-written signature verification, which approaches the problem as a two-class pattern recognition problem. We compute the similarities between the test signature and reference signatures using DTW to get the feature vector of test signature, and then classify it into one of the two classes (genuine or forgery) by the SVM classifier. Compared with other classifiers (Bayesian and Artificial Neural Network), SVM has better classification result on hand-written signature verification. The best result of this algorithm yields FAR of 9.50% and FRR of 13.33%, which has better result than single DWT algorithm with FAR of 18.12% and FRR of 10.33%.