DeepSignature: fine-tuned transfer learning based signature verification system

With the rapid advancement in computer science and information technology, the demand for authentication of a person in different organizations, institutions, banks or online trading, etc. is increasing day by day. Similarly, signature verification and recognition systems are frequently used for forgery and fraud detection. Signature authentication and verification have been important bio-metric traits for many decades. In this article, Convolutional Neural Network (CNN) architectures namely AlexNet, GoogleNet, ResNet, MobileNet, and DenseNet are examined and compared. Further, a fine-tuned transfer learning-based approach is also investigated to verify and identify the offline images of signatures. The evaluation has been done using Persian signatures of different persons using the UTSig dataset as a benchmark. The experimental analyses demonstrate that DenseNet architecture outperformed the other architectures of CNN.