AN ENSEMBLE-DEEP LEARNING FRAMEWORK FOR OFFLINE SIGNATURE VERIFICATION WITH FORGERY ROBUSTNESS

УДК 004.8
DOI: 10.36871/2618-9976.2025.9–2.016

Авторы

Thi Thu Nguyen,
Sao Do University, Chu Van An, Hai Phong, Vietnam
Rashid A. Ishmuratov,
Kazan State Power Engineering University, Kazan, Russia

Аннотация

Offline signature verification is an important problem in the field of security and identity authentication. This paper proposes an ensemble deep-learning framework to improve the accuracy and anti-forgery ability of signature recognition systems. Three deep learning models — convolutional neural network (CNN), Siamese network and fine-tuned ResNet50 — are integrated through a weighted voting mechanism. Preprocessing includes image normalization, denoising and data augmentation to handle sophisticated forgeries. Experiments on the CEDAR, GPDS and a separate hypothetical dataset show that the proposed method achieves 98.7% accuracy, reducing FAR and FRR compared to existing methods.

Ключевые слова

Offline signature authenticity framework,
composite deep-verification architecture,
adaptive skilled-forgery resilience, multimodal feature synthesis for signatures,
robustness-aware signature classifiers,
Python-driven verification pipeline.