Forgery Detection in Academic Certificates Using Hybrid Deep Learning Framework

The increasing circulation of forged academic certificates presents a major challenge for institutions, employers, and verification agencies, as traditional validation processes rely heavily on manual inspection, template familiarity, and subjective judgment. These conventional methods struggle to identify subtle tampering, unauthorized text modifications, replicated seals, and signature alterations. Variations in document formats, image resolutions, and acquisition conditions further reduce reliability. Additionally, existing automated systems often depend on single-model architectures, which lack robustness against diverse manipulation techniques and fail to generalize across multiple certificate templates. The absence of integrated anomaly detection and signature verification components limits the capacity to detect sophisticated editing patterns. To address these difficulties, this work introduces a hybrid deep learning–based certificate validation framework that combines multiple model families to achieve comprehensive and resilient analysis. Convolutional networks such as ResNet50 and EfficientNetB0 provide strong local feature extraction for detecting pixel-level inconsistencies. A Vision Transformer incorporates global contextual reasoning, enabling detection of subtle modifications and watermark disturbances. A convolutional autoencoder performs reconstruction-driven anomaly assessment, capturing unseen or novel manipulation types. A Siamese network adds metric-based verification of signatures and stamps, ensuring accurate similarity assessment. The fusion of these complementary models enhances detection reliability and strengthens decision consistency across varying certificate structures. Key features of this methodology include multi-model hybrid analysis, anomaly-driven authentication, integrated signature similarity assessment, and rule-based decision fusion for robust verification. This unified approach significantly improves the accuracy, adaptability, and trustworthiness of certificate validation systems, addressing the core challenges present in existing manual and automated methods.