A Lightweight Dual-Modality Biometric Classifier for Hand and Iris Images

Strengthening digital system security, especially by shielding consumers from online fraud and illegal account breaches, is the main goal of this thesis. Unauthorized access to private data has grown to be a major worry in today’s connected society, and strong and dependable authentication systems are desperately needed to protect users.A complicated convolutional neural network (CNN) based multimodal biometric identification approach has been presented to overcome this difficulty. Convolutional, pooling, dense layers, and concatenation are used to conduct feature fusion. This method uses restricted and strategic data augmentation and minimum preprocessing to optimize accuracy and resilience by combining several biometric features.With a Test Accuracy of 98.94%, Genuine Acceptance Rate (GAR) of 97.87%, Weighted F1 Score(test set) of 98.87%, and most importantly an Equal Error Rate (EER) of 0.0000, the assessment demonstrates that the model produced extremely encouraging results. The False Acceptance Rate (FAR) and False Rejection Rate (FRR) of the system are identical, therefore EER is a crucial parameter in biometric security. The system’s ability to nearly flawlessly differentiate between authentic users and imposters is demonstrated by its zero EER, which ensures far greater security than many current methods.The methodology is also appropriate for real-time biometric authentication because it takes an average of about 121 milliseconds each step. These findings demonstrate that the suggested approach is a good option for high security applications as it not only improves accuracy but also offers more defense against unauthorized access and unethical intrusions.