Robust iris image segmentation
Peter Wild, Heinz Hofbauer, James M. Ferryman, Andreas Uhl · Institution of Engineering and Technology eBooks · 2017
In this chapter we presented current trends in iris segmentation, summarising the advances in state-of-the-art individual segmentation, shedding light on the pitfalls of NIR vs. VIS iris segmentation and illustrating means to tune existing iris segmentation towards specific datasets. We highlighted different forms of segmentation accuracy assessment and evaluated different approaches to combine segmentation algorithms. To get more stable results we integrated segmentation quality prediction in the fusion process.Tests on ground-truth confirm this positive effect across the board. For the goal of developing more robust iris segmentation techniques, multi-segmentation fusion and quality prediction can be a versatile tool to obtain more stable and more accurate results. However, performing fusion and running multiple segmentation algorithms impacts the processing time. This was not explicitly covered and further research on frame throughput might be necessary for video-based iris segmentation.