An Online Signature Verification System based on Multivariate Autoregressive Modeling and DTW Segmentation

A new online signature verification system based on multivariate autoregressive (MVAR) modeling in combination with a Dynamic Time Warping-based (DTW) segmentation technique is presented in this work. A uniformly spatial-spaced signature sequence is treated as a two element vector sequence (xj, yj). A modified segment-coordinate dynamic time warping algorithm is employed to improve alignment between the signature samples and a master signature reference for the subject writer. Subsequently, a new MVAR model is used to extract coefficients for each segment to construct a feature vector. These vectors are then fed into a Neural Network with a multi-layer perceptron architecture. The performance of the system was evaluated using a testing set of signatures for each writer. The system achieved preliminary accuracies of: 99.9% in a random forgery test, 98% in casual forgery tests, and 96.6% in a skilled forgery test.