Term	Definition
Iris segmentation	The process of isolating the iris from the rest of the eye in an image, including separating it from other structures like the eyelid, eyelashes, and surrounding areas.
Deep Learning (DL)	A subset of machine learning involving neural networks with multiple layers, used for tasks such as image segmentation and classification.
U-Net/U-Net++	Specific deep learning architectures designed for image segmentation tasks. U-Net++ enhances U-Net by adding nested convolutional blocks to improve feature representation.
Cross-dataset performance analysis	Evaluating a model’s ability to perform well on datasets that were not used during its training phase, important for testing model generalization.
Generalization capability	The ability of a model to apply what it has learned from one dataset to new, unseen data from different datasets.
mIoU (Mean Intersection Over Union)	A metric used to evaluate the accuracy of segmentation models by comparing the overlap between predicted and ground truth areas.
F1 Score	A measure of a model’s accuracy that considers both precision and recall, especially in binary classification tasks like pixel-wise segmentation.
NIR (Near-Infrared)	A part of the light spectrum used in iris imaging, typically in the wavelength range of 700 to 900 nanometers (nm), for capturing iris patterns under controlled lighting conditions.
VIS (Visible Spectrum)	The portion of the electromagnetic spectrum visible to the human eye, ranging from 380 to 720 nanometers (nm). While the use of VIS for iris recognition is still an area of ongoing research, it holds potential to reduce hardware complexities associated with NIR collection, making it easier to integrate iris recognition into portable devices such as smartphones.
Benchmarking	The process of comparing different models or algorithms using specific metrics across common datasets to evaluate their performance.
Eyeglash occlusion	A condition where parts of the iris image are blocked by eyeglashes, making segmentation more difficult.
Practicability	A measure of how feasible it is to implement a model in real-world applications, considering factors like inference time and computational requirements.
Inference time	The time a model takes to make a prediction, critical for real-time applications.
Ground truth	Manually labeled data used as a standard or reference for evaluating the performance of algorithms, such as manually segmented iris masks.
OSIRIS	An open-source iris recognition software that provides a complete iris recognition pipeline, including segmentation, feature extraction, and matching.
USIT	An iris recognition software that provides a complete open-source framework for iris segmentation, feature extraction, and matching, often used in biometric research alongside OSIRIS.
Composite dataset	A dataset created by combining samples from multiple publicly available datasets to form a larger, more diverse set of images for training or evaluation.
Noise detection capability	The ability of a segmentation algorithm to identify and handle artifacts like reflections or occlusions that can interfere with the accuracy of the iris segmentation process.
Specular reflection	The bright spot or glare in an image caused by light reflecting off curved surfaces, such as the eye or eyeglasses. This reflection can obscure iris details and negatively impact segmentation accuracy.
Dilation	The process by which the pupil expands in response to low light or other stimuli. Pupil dilation can alter the visible area of the iris, potentially impacting segmentation accuracy.