IRIS and Face-Based Multimodal Biometrics Systems

Deep learning is a powerful visual model of machine learning tools. A multimodal biometrics system employing deep learning techniques with Face and Iris attributes improves human recognition’s security level and accuracy. The proposed work plans to develop a multimodal biometrics system using Face and Iris traits by identifying an optimal fusion level and efficient fusion rule for secure person authentication. In the proposed work, transfer learning methods, namely ResNet-50 and VGG16, are used for extracting features, and Softmax layer and Multi-class Support Vector Machine are used for personal identification. The performance of the proposed system is evaluated using the VISA multimodal biometrics database. The accuracy of 93.334 and 96.735% are achieved for feature and score-level fusion.