Biometric Fusion Approaches Based on Deep Convolutional Neural Network
Hawra Ali Hussain, Hawraa H. Abbas · 2023
Spoofing attacks have increased significantly, and a lot of research is being done on biometric security systems. Researchers are more interested in multimodal biometrics to give better security using biometric applications. A complicated model structure with minimal risk of spoofing attack can be created using multimodal biometrics applications. The new multimodal human identification model presented in this paper was created by fusing the two multimodal biometrics methodologies (face, iris, and fingerprint). The Convolutional Neural Networks (CNNs) that form the system’s structure extract features from the photos and use the Softmax classifier to categorize them. Five different CNN algorithms (VggNetl6, ReseNet50, MobilevNet, DenseNet, GoogleNet) are used with two fusion approaches (score and feature level). The designed automatic identification system is conducted on the SDUMLA-HMT dataset, the best-acquired accuracy results are 99.37% for score-level fusion and the accuracy is 97.55% with a feature-level fusion approach when the VggNetl6 algorithm is used. This system’s accuracy results are outperforming the latest current method methods.