Non-segmentation and Deep-Learning Frameworks for Iris Recognition.

Traditional iris recognition algorithms and data-driven iris recognition generally believe that the iris recognition process should be divided into a series of sub-processes which makes the iris recognition process more complex. Furthermore, each sub-process relies on specific algorithm which greatly increases the computational complexity of the overall framework of the model. The work proposes an end-to-end iris recognition algorithm based on deep learning. We encourage the reuse of feature and increase the interaction information across channels with the aim of improving the accuracy and robustness of iris recognition, at the same time, to some extent, reducing the complexity of the model. Related experiments are conducted in three publicly available databases, CASIA-V4 Lamp, CASIA-V4 Thousand and IITD. The results showed that non-process-based iris algorithm that we proposed consistently outperforms than several classical and advance iris recognition methods.