摘要
在实际环境中,虹膜图像的采集经常会遇到被眼睑或睫毛遮挡等特殊情况,不利于虹膜的分割和定位。提出一种改进的卷积神经网络实现对虹膜部分的提取和定位。组合多种设备和环境下采集的虹膜数据,基于MobileNet对原始Unet的编码层进行修改,并与原方法进行比较,实验表明加入深度可分离卷积模块后的分割模型可以更准确、更快速地分割虹膜,对特殊情况下采集的虹膜图像具有很好的鲁棒性。
In the real environment,the collection of iris images often encounters special circumstances such as obscuration by the eyelids or eyelash⁃es,which is not conducive to the segmentation and positioning of the iris.This paper proposes an improved convolutional neural network to extract and locate the iris part.Combining iris data collected under multiple devices and environments,and modifying the original Unet coding layer based on MobileNet and comparing it with the original method.Experiments show that the improved segmentation model can segment the iris more accurately and faster.The iris image acquired under the image is very robust.
作者
于泰峰
YU Tai-feng(College of Computer Science,Sichuan University,Chengdu 610065)
出处
《现代计算机》
2020年第15期121-125,共5页
Modern Computer
关键词
语义分割
Unet
虹膜
深度可分离卷积
Semantic Segmentation
Unet
Iris
Depthwise Separable Convolution
