Iris Recognition based on Convolutional Neural Network and Active Iterative optimization

Iris recognition has become a popular method for authentication due to its high accuracy, non-invasiveness, and resistance to spoof attacks. Deep learning techniques have significantly enhanced the precision and effectiveness of iris recognition, but there are still three main problems: (1) iris images often have occlusion and noise; (2) there is a lack of effective iris samples, and the classification categories are numerous, making it difficult to effectively utilize the difficult-to-distinguish samples; (3) the model is not lightweight enough for practical applications. Considering these challenges, This paper introduces a robust and lightweight iris recognition approach that leverages convolutional neural networks and draws inspiration from Active Learning to enhance optimization. Firstly, the iris images are preprocessed using a series of image enhancement techniques. Then a lightweight classifier improved MobileNetV2 is used to extract features for recognition. Meanwhile, transformer learning is added to initialize the classification network parameters. At last, the active iterative optimization method is adopted in the training process to fully exploit the sample information. The experimental results clearly demonstrate that the proposed method surpasses several state-of-the-art iris recognition techniques in terms of accuracy, recall, and F1-score.