Survey Available Techniques for Signature Fraud Detection Using DL Algorithms

Signature fraud poses a persistent challenge across various domains, from financial transactions to legal documents, demanding robust and reliable detection methods. This paper presents a comprehensive review of existing signature fraud detection techniques, specifically focusing on approaches leveraging deep learning (DL) algorithms. The study investigates a range of DL architectures, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Siamese networks, analyzing their effectiveness in extracting discriminative features from signature images or dynamic signature data. We categorize and compare approaches based on data preprocessing, feature extraction techniques, model architectures, and performance metrics. The review also explores publicly available signature datasets and their suitability for training and evaluating DL models. Finally, we identify key challenges and limitations in the current state-of-the-art, including the impact of skilled forgeries, variations in signature styles, and the need for robust generalization capabilities. The survey concludes by highlighting promising directions for future research, such as developing explainable AI (XAI) techniques to enhance trust and transparency in signature fraud detection systems and exploring the potential of adversarial training to improve model robustness against sophisticated attacks.