On-Line Authenticity Verification of a Biometric Signature Using Dynamic Time Warping Method and Neural Networks

Krzysztof Walentukiewicz, Albert Masiak, Aleksandra Galka, Justyna Jelinska, Michal Lech

Abstract: To ensure proper authentication, e.g. in banking systems, multimodal verification are becoming more prevalent. In this paper an on-line signature based on dynamic time warping (DTW) coupled with neural networks has been proposed. The goal of this research was to test if combining neural networks with DTW improves the effectiveness of verification of a handwritten signature, compared to the classifier based on fixed thresholds. The DTW algorithm was used as a feature extraction method and a similarity measure. Two neural network architectures were tested: multilayer perceptron (MLP) and one with convolutional neural network (CNN). A dataset containing model, verification and forged signatures gathered from a research group using a biometric pen has been created. The research has proved that the DTW coupled with neural networks perform significantly better than the baseline method - DTW model based on constant thresholds. The results are presented and discussed in this paper.