An End to End Deep Neural Network for Iris Recognition

Abstract

Abstract The application of biometrics technology in all areas of people’s lives in today’s intelligent era. With the advantages of high accuracy and contactless, iris recognition is an important and challenging research area. In this work, an application of the combined network model based on EfficinetNet-b0 is presented in iris recognition, which integrates iris segmentation, normalization, iris feature extraction and matching into a unified network. The network model has high parameter efficiency and speed. Compared with previous deep iris recognition network, the network architecture has three characteristics: (1) Compared with most existing training and phase adjustment algorithms, it is end-to-end trainable. (2) Grad-cam has class recognition and high resolution. It provides a good visual interpretation. (3) An effective and smaller baseline model is proposed that balances the depth, width and resolution of the network based on the scaling model and achieves better results. The hybrid iris databases, composed of CASIA Thousand and Mmu2, proves that the accuracy and efficiency of the composite network framework are better than those of the previous network framework. The visualization of data sets is validated, which proves that the combined model is robust to iris image localization.