Machine Learning and Dynamic Time Warping for Online Signature Verification

Signature verification is a very common task in the field of document analysis used for the identification and validation of individuals. Many approaches can be used for signature classification. Using a machine-learning-based system is one of the most common approaches. This paper uses machine learning to classify signatures as genuine or forged. It also used a dynamic time-warping approach in the preprocessing step to produce data for training purposes. This approach differs from using it in the classification phase, which is a common practice in signature verification. The work targets online signatures, which are obtained using digital devices. Different databases were used to evaluate the accuracy of the verification systems, and four different algorithms were used for classification. The results are very promising. The minimum error rate achieved was 2.09% for the MCYT-100 dataset, 3.34% for the SVC2004 dataset, $\mathbf{1. 6 8 \%}$ for the SigComp’11-Chinese dataset, and $\mathbf{3. 6 8 \%}$ for the SigComp’11- Dutch dataset.