Faceirisnet: A Deep Multimodal Biometric Recognition Framework Using Resnet And Triplet Loss

Abstract

Facial and Iris recognition are essential parts of biometric security systems used to ensure reliable identification and authentication in high-stakes situations. For accurate facial and iris identification tasks, this study proposes a dual Convolutional Neural Networks (CNNs)-based architecture. The integrated system offers a multi-modal approach to biometric verification by processing iris and facial images concurrently. The CNN architecture used by the facial recognition module is based on ResNet and uses residual connections and deep feature extraction to solve the vanishing gradient issue. By optimizing a pre-trained ResNet model on the VGGFace2 dataset, transfer learning is used to achieve high accuracy in face identification and verification in difficult circumstances like occlusion and lighting changes. In order to separate the iris from eye pictures, the iris identification module uses a modified ResNet model that is tuned for fine-grained iris textures and incorporates a sophisticated segmentation technique. Model resilience is improved by data augmentation methods like rotations and random cropping. A normalized iris dataset is used to train the CNN, allowing for the extraction of discriminative iris characteristics that are necessary for accurate identification. A fully connected layer performs final classification after a late-fusion approach concatenates embeddings from both CNNs to merge facial and iris data for safe authentication is FACEIRISNET. It builds a strong multi-modal biometric system by utilizing both facial and iris features. Both facial and iris embeddings use a triplet loss function, which makes sure that embeddings of the same identity are grouped together and those of different identities are pushed away.