基于U-Net3+的虹膜分割算法研究  

Research on Iris Segmentation Algorithm Based on UNet3+

作　　者：张星星 娄莉[1] ZHANG Xingxing;LOU Li(Xi’an Shiyou University,Xi’an Shaanxi 710065,China)

机构地区：[1]西安石油大学,陕西西安710065

出　　处：《信息与电脑》2024年第17期1-5,共5页Information & Computer

摘　　要：虹膜分割是虹膜识别中非常重要的环节,会影响最终识别的结果。针对虹膜图像中存在的噪声问题,本文使用了图像分割网络结构U-Net3+,来进行虹膜分割。相较于U-Net,U-Net3+网络结构增加了深度监督和全尺度跳跃连接,提高了虹膜分割的准确性。本文在U-Net3+中引入了CA(Coordinate Attention)注意力机制模块,使网络结构更加关注重要的特征部分,并将模型在CASIA-Iris数据集上进行了验证。实验结果表明,U-Net3+和U-Net3+(CA)有更高的精度。The iris segmentation is a very important step in iris recognition,which will affect the final recognition result.Aiming at the noise problem in iris images,this paper uses an image segmentation network structure U-Net3+to perform iris segmentation.Compared with U-Net,U-Net3+network structure adds depth supervision and full-scale jump connection to improve the accuracy of iris segmentation.In this paper,the CA(Coordinate Attention)attention mechanism module is introduced into U-Net3+to make the network structure pay more attention to the important feature parts,and the model is validated on the CASIA-Iris dataset.The experimental results show that U-Net3+and U-Net3+(CA)have higher accuracy.

关 键 词：虹膜分割 U-Net3+ 虹膜识别 CA注意力机制 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]