MACHINE LEARNING-BASED OFFLINE HANDWRITTEN SIGNATURE VERIFICATION: A SURVEY AND PROPOSED FRAMEWORK

Offline handwritten signature verification remains an important biometric authentication problem in applications such as banking, legal documentation, access control, and identity validation. Manual verification is often time-consuming, subjective, and vulnerable to error, particularly in the presence of skilled forgeries. This paper presents a survey of machine learning approaches for offline handwritten signature verification and proposes a general framework for designing such systems. The paper reviews the main stages of the verification pipeline, including image pre-processing, feature extraction, classification, and threshold-based decision making. Both handcrafted and learned feature representations are discussed, with particular attention to convolutional neural networks, support vector machines, deep metric learning, and Siamese architectures. Common benchmark datasets and evaluation measures, including accuracy, precision, recall, F1-score, false acceptance rate, false rejection rate, and equal error rate, are also summarized. Based on the reviewed literature, a modular verification framework is proposed to support future research and practical implementation. The paper further highlights key challenges, including intra-class variation, skilled forgery detection, limited training data, and generalization across writers, and outlines promising directions for future work in secure and scalable signature verification systems.