Performance Analysis of Deep Neural Networks for Unimodal and Multimodal Biometric Authentication

Abstract

Biometric technology is a powerful tool that relies on distinct and measurable physiological or behavioral characteristics possessed by individuals. These traits serve as reliable means to verify and authenticate individuals. However, unimodal biometric systems encounter several challenges that hinder their effectiveness. These challenges include noisy data, variations within the same class, limited degrees of freedom, non-universality, susceptibility to spoof attacks, and high error rates. To address these challenges, researchers have turned to multimodal biometric systems. These systems leverage the use of two or more biometric modalities to enhance performance and overcome the limitations of unimodal systems. In the context of the discussed system, face and fingerprint modalities are utilized. To develop and evaluate the performance of the system, various deep learning algorithms such as VGG16, VGG19, CNN, and Inception are employed. These algorithms are trained, validated, and tested using the face and fingerprint data. Performance metrics such as accuracy, precision, and f1 score are calculated for both unimodal and multimodal configurations. The results of the performance evaluation demonstrate that CNN (Convolutional Neural Network) yields higher accuracy compared to the other models tested. This finding suggests that CNN is particularly well-suited for this biometric system, as it can effectively extract meaningful features from the face and fingerprint data and provide accurate classification results. By adopting a multimodal approach and utilizing deep learning algorithms such as CNN, the system successfully addresses the challenges faced by unimodal biometric systems. This implementation demonstrates improved accuracy, making it a promising solution for reliable and secure individual authentication in various applications.