Study on iris segmentation algorithm based on dense u-net

Abstract

Iris segmentation is an essential process of iris recognition. Iris segmentation plays an important role in maintaining the accuracy of iris based on recognition system by limiting the errors in the current stage. However, its performance is affected by non-ideal conditions caused by ambient light noise and user non-cooperation. The existing segmentation methods based on local features cannot find the real iris boundary under these non-ideal conditions, and the errors generated in the segmentation stage will traverse to all subsequent stages, resulting in decreased accuracy and reliability. In addition, real iris boundaries need to be divided without additional denoising costs. Aiming at the problems of existing algorithms in complex scenes and cross-device applications, an Iris segmentation algorithm based on Dense U-Net is presented in this paper. Combining Dense network with U-Net network, Iris is segmented by taking advantage of dense U-Net network, which is narrower and has fewer parameters, and taking advantage of U-Net in semantic segmentation. Dense connected path is derived from dense connected network (Dense U-Net), in which improved information and parameters are helpful to reduce the training difficulty of deep network. The final segmentation accuracy was 98. 36%. F1 is 97.07%. The experimental results prove the presented model can improve the accuracy, reduce the error rate, and assist doctors in the diagnosis of Iris Diseases effectively.