Beyond the Pen: Deep Learning Advances in Offline Signature-Based Writer Identification and Verification

Writer identification and verification have emerged as critical tasks in biometric authentication systems, with offline handwritten signatures remaining one of the most widely accepted forms of identity verification. This paper presents a comprehensive review of recent advancements in deep learning approaches tailored for writer identification and verification, focusing on offline signature analysis. We explore the transition from traditional feature engineering methods to data-driven models powered by Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and more recently, Transformer-based architectures. The paper discusses the strengths and limitations of key model types, including Siamese and Triplet networks, in capturing writer-specific traits and distinguishing between genuine signatures and skilled forgeries. Furthermore, we evaluate training strategies, loss functions, and the role of transfer learning in enhancing model generalizability across datasets. Key benchmark datasets such as GPDS, CEDAR, and MCYT are reviewed to highlight challenges in standardization and cross-domain performance. Finally, the paper outlines open research problems, including data scarcity, explainability, and real-world deployment constraints, providing directions for future research in robust and scalable writer verification systems.