EXPLORE PUBLICATIONS BY COUNTRIES

Tamper Detection, Deep Learning, Traditional Image Processing, Siamese Network, Identity Verification, Image Analysis.

Ensuring the integrity and security of identity cards to prevent fraud also to conservate institutional credibility in educational institutions is crucial. This study presents a comparative analysis of traditional image processing techniques and deep learning methods for tamper detection in Nigerian university system identity cards. Traditional methods evaluated include Canny edge detection, histogram comparison, Sobel edge detection, and Laplacian edge detection, while the deep learning method uses a Siamese Network. The dataset, composed of original and tampered identity cards which were generated from original identity cards through blurring, noise addition, shifting, and text alterations, was evaluated using accuracy, precision, recall, F1-score, and ROC AUC metrics. From the results of this study, it was shown that the Siamese network achieved the highest accuracy (80%) with an F1-score of 0.89, while Canny edge detection followed closely with an accuracy of 79% and F1-score of 0.88. Other traditional methods such as Sobel, Laplacian, and Histogram comparison underperformed, achieving accuracies below 30%. The results show that the Siamese network is more effective in detecting subtle tampering and generalizes better on limited datasets compared to traditional methods. Finally, this study concludes that deep learning, specifically the Siamese Network, provides superior accuracy and reliability, making it a more effective solution for tamper detection in Nigerian university identity systems.