Offline Signature Verification: An Extensive Survey of Deep Learning Methods
Gaurav Kumar Pandey, Vavilala Divya Raj, Ayush Agarwal, Mayank Dixit, Sachin Singh Chauhan, Sumit Srivastava · 2025
Offline signature verification is essential for identity authentication and fraud mitigation in many applications, such as banking and legal documentation. Conventional approaches to signature verification predominantly depended on manually designed features and statistical methodologies, although they frequently encountered difficulties with intricate intra-class variances and inter-class similarities. In recent years, deep learning has arisen as a potent instrument to address these difficulties through autonomous feature extraction and resilient categorization. This study offers an extensive examination of cutting-edge deep learning methodologies for offline signature verification. We examine fundamental designs like Convolutional Neural Networks (CNNs), hybrid models, and transfer learning methodologies, and evaluate their efficacy on renowned signature datasets such as CEDAR and GPDS. Furthermore, we examine the difficulties related to imbalanced datasets, counterfeit detection, and the choice of suitable assessment metrics, including accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR). This paper identifies existing limits in DL methodologies and proposes future research routes to tackle challenges including dataset scarcity, model generalization, and computational efficiency. Our findings provide studies with insights into the advantages and disadvantages of diverse deep learning methodologies, thereby aiding in advancing more precise and efficient signature verification systems.