A hybrid cnn with vision transformer technique for effective iris recognition

Abstract

Biometric reputation has experienced an extensive increase, with many businesses integrating biometric technology into their structures. Among those, iris recognition stands out due to its effectiveness in preventing identity fraud through disposing of the risks of collisions or fake suits, even with large datasets. While Convolutional Neural Networks (CNNs) have shown excessive accuracy, they require great datasets and computational assets. To cope with this, this observer proposes a hybrid version that mixes CNNs and Vision Transformer (ViT) for efficient iris picture identification and authentication. By optimizing the knowledge of price, the hybrid model achieves an impressive accuracy of 99.67% in iris recognition. The use of a cross-entropy loss function minimizes prediction mistakes and enhances class labeling accuracy. Additionally, the research introduces a progressive neural network-based total prediction model, Interleaved Residual (IRU-Net), for semantic segmentation and iris mask generation, alongside the Predicted ID model for precise identity predictions. The proposed version undergoes rigorous testing on three extensively available iris databases, demonstrating robust reputation performance. Moreover, the model indicates ability packages in different biometric fields, which include face and retinal reputation.