Privacy‐Centric Signature Classification for Secure Educational Certificate Authentication

ABSTRACT This study aims to develop an efficient deep‐learning‐based approach for the classification of signatures on educational certificates, focusing on enhancing verification processes to ensure the authenticity and integrity of academic credentials. Signature verification, as a critical biometric tool, plays a vital role in combating fraud and maintaining trust in academic environments. To address the challenges of distinguishing between genuine and forged signatures, we utilized a comprehensive dataset comprising authentic and fraudulent samples. The proposed methodology incorporates preprocessing techniques, such as bilateral filtering, to improve image quality, followed by feature extraction using advanced convolutional neural networks (CNNs). Our approach includes a comparative evaluation of multiple baseline architectures, including MobileNetV2, LeNet, and BiOpt‐SVF a CNN‐based model specifically designed for this task. Experimental results reveal that the proposed model achieves superior performance, with accuracy rates of 99.14% on the CEDAR dataset and 98.98% on the GPDS‐150 dataset, demonstrating its effectiveness and reliability in signature verification for academic applications.