Multimodal biometrics has recently gained interest over single biometric modalities. This interest stems from the fact that this technique offers improvements in recognition and more security. In this ongoing research programme, we propose a new feature extraction technique for a biometric system based on face and iris recognition. The extraction of iris and facial features is performed using the Discrete Wavelet Transform combined with the Singular Value Decomposition. Merging the relevant characteristics of the two modalities is used to create a pattern for each individual in the dataset. The evaluation process is performed using two datasets (i.e., Faces94 Faces dataset and IIT Delhi Iris dataset). The experimental results carried out in this programme showed the robustness of the proposed technique.

Biometrics is an alternative solution to the old means of identity verification, such as access cards. However, monomodal biometric systems suffer from multiple limitations, such as the noise introduced by the sensor and non-universality. Multi-biometrics allows us to overcome these problems and thus obtain better performance. Multimodal biometrics, using deep learning, has recently gained interest over single biometric modalities. In this work, we propose a deep learning model for persons identification/verification using Face and Iris traits. The features of the iris and the face are extracted utilizing the DenseNet121 and FaceNet models, and these features are merged using a feature-level fusion scheme. We also proposed a new automatic matching technique to verify the person’s identity. The results presented in this paper show interest in deep learning approaches for face and iris recognition, especially when models are pre-trained. The results also highlighted the interest in the proposed DenseNet121-FaceNet model and the automatic matching method compared to the standard threshold selection according to the Equal Error Rate point.

No abstract