A behavioral biometric is not based on the individual's physical properties, such as fingerprints or faces, but behavioral ones. Every person has a distinctive signature, primarily used for personal identification and to confirm the authenticity of important papers or legal transactions. In biometric authentication, signature verification is an important study subject. The project aims to create a personal signature-based authentication system. In this work, data extracted from the ICDAR dataset are used, which contains the signatures of Dutch users, both genuine and fraudulent. The data was obtained from the Kaggle website. Two different convolutional neural network strategies have been used to build the proposed model. In the first strategy, the convolutional neural network model was built from scratch; in the second strategy, the pre-trained model, VGG16, was utilized to classify genuine and forged signatures. The findings show that the results of the VGG16 model represent the optimal model for signature forgery verification with an accuracy of 99.8 %, precision of 100%, recall 99.5%, F1-score of 99.4%, and training time of 18 min 52 s

Handwritten Signature; Offline Signature; Convolutional Neural Networks; Deep Learning