Dual-Model Deep Learning Approach for Urdu and English CNIC Authentication: A Robust Solution against Identity Fraud

Abstract

Targeting both Urdu and English Computerized National identification Cards (CNICs), CNIC Authenticity Checker is a potent deep learning-based solution designed to tackle the serious problem of fake identification documents. This study presents a dual-model technique that can accurately recognize false CNICs across both Urdu and English card types, in contrast to existing systems that frequently concentrate on a single language format. Two distinct deep learning architectures were used in order to efficiently handle the structural and design variations between Urdu and English CNICs. The DenseNet201 model's deep and effective feature extraction capacity was used to train it to categorize Urdu CNICs. Three different models were constructed to support different English CNIC subclasses, while the EfficientNetV2B3 model was simultaneously improved to handle English CNICs. Using data augmentation, a weighted binary cross-entropy loss function, and important performance indicators like accuracy, precision, recall, AUC, and confusion matrix visualization, each model was rigorously trained and evaluated. This manual routing preserves simplicity and versatility while guaranteeing precise classification. The system architecture makes the solution feasible for real-world implementation by facilitating smooth integration and offering an intuitive user experience.