Fast Iris Segmentation under Partly Occlusion Based on MTCNN and Weighted FCN

Abstract

Many times, an ophthalmologist will infer the health of the eye, the development of eye diseases, and the recovery by observing the morphological changes of the iris tissue. Therefore, accurate and automatic segmentation of the iris is a very important task. In this paper, we propose an iris segmentation method to tackle with the partly occlusion case that includes fast eye detection based on MTCNN, iris segmentation based on Weighted FCN and Hough Transform and coordinate correction for radius of iris in the real world. Firstly, we apply Multi-task Cascaded Convolutional Networks for eye detection, which is light and fast. Then we propose Weighted FCN and Hough Transform to segment the iris, even if the iris is partially occlusive. Finally, we design a calibration scheme to correct the iris radius in the real world. Experimental results show that the accuracy rate of the proposed method reaches 97.6% and precision rate 98.5%, superior to state-of-the-art methods.