Iris segmentation in uncooperative and unconstrained environments: State-of-the-art, datasets and future research directions

Most of the classical iris recognition systems require cooperation from users and assume ideal conditions during iris image acquisition. In uncooperative and unconstrained environments, the performance of these classical systems is significantly degraded due to the existence of various noise factors such as high occlusions, specular reflections, motion blur, lighting variations and off-angle images. Iris segmentation step has a great significance in the iris recognition system since its errors directly propagate to the next processing steps and affect the iris template as well as the recognition results. Iris segmentation step becomes harder and more challenging in uncooperative and unconstrained environments since most of those noise factors should be handled in this step. Therefore, in recent years the iris segmentation in unconstrained environments has received more attention from the scientific community. However, the state-of-the-art related to iris segmentation in unconstrained environments and its datasets is not well known. In this paper, we aim to shed the light on the most recent advances in iris segmentation in uncooperative and unconstrained environments. Furthermore, the current publicly available iris datasets and their ground truth masks that are commonly used for training or testing segmentation algorithms are examined. The limitations of existing methods and the future research directions are discussed and highlighted to assist researchers in filling the research gap for this problem domain.