Robust Iris Code Generation Using Segmentation Networks for Off-Angle Iris Recognition

Standoff iris recognition systems, which focus on the recognition of non-cooperative subjects in less constrained environments, face significant challenges due to the potential for non-ideal images, particularly those captured from off-angle perspectives. These images are further complicated by factors such as corneal refraction and limbus occlusion. Iris biometrics, which converts unwrapped iris textures into binary iris codes, relies on comparing these codes by measuring their Hamming distances. However, the distortions introduced by off-angle imaging often led to inconsistencies in iris code generation, thereby reducing the accuracy and robustness of recognition systems. This paper introduces a novel approach for robust iris code generation using deep segmentation networks to address these inconsistencies in off-angle iris images. The method processes normalized iris images as input and employs a DeepLab segmentation network to generate consistent iris codes across varying angles for the same subject. During training, both frontal and off-angle normalized iris images are paired with target iris codes derived from the subject's frontal iris code. The segmentation network is trained to minimize the divergence between the generated iris codes and the target codes, ensuring consistent code generation regardless of the image's angle. Experimental results demonstrate that the proposed method significantly improves the consistency and accuracy of iris code generation for off-angle images, outperforming traditional iris recognition techniques. Notably, the average genuine Hamming distance score was reduced by 25 %, and the equal error rate was lowered by 53 %, showcasing the effectiveness of the approach. These findings highlight the potential of segmentation networks to enhance the robustness and reliability of iris recognition systems, particularly in standoff biometric applications.