Development of a Model to Enhance the Security of Signature Verification Systems Using Deep Learning Algorithms (MobileNetV2/LSTM)

Abstract

The aim of this paper was to develop a hybrid model to improve the efficiency and reliability of Offline Arabic Electronic Signature Verification by integrating MobileNetV2 for spatial feature extraction with an LSTM network for analyzing the temporal characteristics of signatures. The study seeks to determine the effectiveness of the proposed model in improving the accuracy of Arabic signature verification and its ability to reduce errors compared with traditional models.

The model was trained and evaluated on an Arabic signature dataset. The experimental results showed that the proposed MobileNetV2-LSTM-Attention-Security model achieved high performance in Arabic signature verification, with an Accuracy of 98.03%, Precision of 98.38%, Recall of 97.66%, F1-Score of 98.02%, and AUC of 0.9925. It also achieved low error rates, with FAR = 1.61%, FRR = 2.34%, and EER ≈ 1.90%, demonstrating its strong ability to distinguish between genuine and forged signatures.