Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{190744,
        author = {Dr. Annapurna H},
        title = {Deep Neural Networks in Offline Handwritten Signature Verification: A Systematic Review (2016–2025)},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {8},
        pages = {1852-1858},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=190744},
        abstract = {Offline handwritten signature verification remains a challenging biometric problem due to high intra-writer variability, low inter-writer variability, and the absence of dynamic writing information. Over the past decade, deep neural networks have significantly advanced this field by enabling the automatic learning of features and the robust representation of complex signature patterns. This paper presents a comprehensive systematic review of deep learning-based approaches for offline handwritten signature verification published between 2016 and 2025.
The review analyzes a wide range of deep architectures, including convolutional neural networks, Siamese and metric-learning frameworks, recurrent models, attention-based networks, and transformer-based architectures. Emphasis is placed on how these models address key challenges such as limited training data, skilled forgery detection, cross-writer generalization, and variability across writing styles and scripts. Commonly used benchmark datasets, including GPDS, MCYT, CEDAR, BHSig260, and other multilingual signature corpora, are examined in terms of their role in performance evaluation and comparability across studies.
The surveyed works are systematically categorized according to architectural design, learning strategy, and evaluation protocol, highlighting evolving research trends and methodological shifts over time. Key findings indicate a transition from conventional convolutional pipelines toward hybrid and attention-based models that capture both local stroke-level details and global structural dependencies. Despite significant progress, challenges such as dataset imbalance, limited generalization across writing styles, and lack of standardized evaluation protocols remain open research problems.
This review presents a comprehensive overview of the current state of the art in offline handwritten signature verification, identifying promising research directions, including self-supervised learning, foundation models, and cross-domain adaptation, to support the development of more robust and scalable verification systems.},
        keywords = {Offline handwritten signature verification, Deep neural networks, Convolutional neural networks, Siamese networks, Transformer models, Deep metric learning, Biometric authentication, Signature datasets, Writer-independent verification, Pattern recognition.},
        month = {January},
        }