Machine learning approaches for accurate iris segmentation

Iris segmentation is an essential problem area that has attracted researchers interest during the years. Previous approaches to iris segmentation have tended to be hand-tuned methods that, however, provided high segmentation accuracy despite the images obtained in low-light conditions. Since successful employment of deep learning models, researchers are turning their attention towards utilization of convolution neural networks (CNN) for the enhanced improvement in the existing iris segmentation, and several CNN based technique has already emerged into the literature recently. Earlier we had employed thresholding techniques, Hough transform, Daugman transform for segmentation of iris region. However, the results were not up to our expectation; thus, we adopted deep learning methods for iris segmentation. In this paper, we have employed U-Net architecture, which is present in nearly all fields of image segmentation.