Deep Feature Extraction Technique Based Multimodal Biometric Identification Approach
Padmaja Pulicherla, Zaid Ajzan Alsalami, S. Divya, Vankudoth Ramesh, Illa Mahesh Kumar Swamy · 2024
In modern times, providing accurate information security has become a difficult task with the increase in technology of cyber-attacks. One of the most convenient and secure way of authenticating is fingerprint but the single modal biometric identification is at risk. Employing multimodal biometric recognition systems provides efficient security in identifying authentic users. This research provides a multimodal biometric system using face and finger vein by employing Residual Network and Long Short-Term Memory (ResNet-LSTM) method. The preprocessing of input images is performed by Contrast Limited Adaptive Histogram Equalization (CLAHE) technique to enhance local contrast of image. Residual Network (ResNet) with residual blocks is utilized for spatial hierarchical feature extraction and Long Short-Term Memory (LSTM) for temporal extraction, retaining necessary information. In feature fusion, Self-Attention mechanism is employed to determine relationship among different elements and Softmax layer to convert weights into probabilities of classes. The performance of ResNet-LSTM method is evaluated on SDUMLA-HMT dataset and achieved 99.92% accuracy. The proposed method outperformed existing biometric recognition systems like Political Optimizer with Deep Transfer Learning Enabled Biometric Iris Recognition (PODTL-BIR), Convolutional Neural Network (CNN).