Fake Signature Recognition Using Convolutional Neural Network with ResNet-18 Architecture

Abdul Aziz Al Basyir
Kenan Almas
Big Daya Yudha Asmara

Frontier Advances in Applied Science and Engineering
Vol. 3 No. 2 (2025)

Signatures are widely used as a personal identity marker and as a key element in validating official documents. Although each individual’s signature is generally distinctive, genuine signatures may vary in shape and size across different instances, while forged signatures can appear visually similar to authentic ones. These factors make manual verification subjective, time-consuming, and prone to error, particularly when handling large volumes of documents. This study developed an automatic signature recognition system using deep learning, specifically a Convolutional Neural Network (CNN) with a ResNet-18 architecture. The dataset consisted of 1,038 signature images, and model performance was evaluated across multiple training epochs. The best result was obtained using the Adam optimizer at epoch 25, achieving an accuracy of 59.5% on the evaluated data. The findings indicate that high visual similarity among signature images can substantially affect classification outcomes, suggesting that offline signature recognition remains challenging when formulated as a simple binary image classification task. Future work should consider improving the training protocol and evaluation scheme to better capture subtle discriminative features between genuine and forged signatures.