Unsupervised iris image segmentation

Abstract

Iris biometrics is considered one of the most accessible and effective biometric characteristics due to its unique patterns and stability over the time. This paper presents a neural network method based on W-Net architecture for iris image segmentation. The use of Daugman’s integro-differential operator enhances the precision and effectiveness of iris image segmentation. The W-Net architecture, utilizing deep learning, accurately isolates the iris region, while the Daugman’s operator ensures robust boundary detection. The model is trained on the CASIA-IrisV4-Interval dataset and effectively handles challenges such as eyelash occlusions. This method opens new possibilities for developing more reliable and accurate biometric identification systems with applications in access control and identity verification.