Boosting Iris Recognition by Margin-Based Loss Functions

In recent years, the topic of contactless biometric identification has gained considerable traction due to COVID-19 pandemic. One most well-known technologies is iris recognition. Determining classification threshold for large datasets images remains challenging. To solve this issue, it essential extract more discriminatory features from images. Choosing appropriate loss function enhance discrimination power one significant factors in deep learning networks. This paper proposes a novel framework that integrates light-weight MobileNet architecture with customized ArcFace and Triplet functions. By combining two functions, possible improve compactness within class discrepancies between classes. reduce amount preprocessing, normalization step omitted segmented are used directly. contrast original SoftMax loss, EER combined decreased 1.11% 0.45%, TPR increased 99.77% 100%. CASIA-Iris-Thousand, 4.8% 1.87%, while improved 97.42% 99.66%. Experiments have demonstrated proposed approach using can significantly state-of-the-art achieve outstanding results.