光照变化条件下基于深度学习的人脸轮廓识别  

Face contour recognition based on depth learning under the condition of lighting changes

作　　者：李益沛[1] 宋鹏波 李晶晔 章玉龙 李杉懋 LI Yipei;SONG Pengbo;LI Jingye;ZHANG Yulong;LI Shanmao(State Grid Shanxi Electric Power Company,Taiyuan 030025,China;Information and Communication Branch of State Grid Shanxi Electric Power Company,Taiyuan 030025,China;State Grid ICT Industry Group Anhui Jiyuan Software Co.,Ltd.,Hefei 230088,China)

机构地区：[1]国网山西省电力公司,山西太原030025 [2]国网山西省电力公司信息通信分公司,山西太原030025 [3]国网信通产业集团安徽继远软件有限公司,安徽合肥230088

出　　处：《电子设计工程》2024年第13期181-184,189,共5页Electronic Design Engineering

摘　　要：在光照不均匀变化条件下难以精准区分人脸轮廓,导致人脸识别精度较低,因此提出光照变化条件下基于深度学习的人脸轮廓识别方法。在光照变化条件下,采集人脸轮廓三维图像信息。使用深度学习方法提取人脸轮廓特征,计算训练集字典稀疏程度,完成人脸轮廓输入数据学习与训练。引入二阶梯度信息,进行人脸轮廓频谱匹配。使用高斯统计去除离群点,区分光线和人脸轮廓,获取完整人脸轮廓。实验结果表明,在不同光照条件下,该方法的最高人脸轮廓识别精度为96%,因此说明该方法具有较高的识别精度。It is difficult to accurately distinguish the face contour under the condition of uneven lighting changes,which leads to low face recognition accuracy.Therefore,a face contour recognition method based on depth learning under the condition of lighting changes is proposed.Under the condition of lighting changes,the 3D image information of face contour is collected.The depth learning method is used to extract the face contour features,calculate the sparse degree of the training set dictionary,and complete the face contour input data learning and training.The second step information is introduced to match the face contour spectrum.Gaussian statistics is used to remove outliers,distinguish light and face contour,and obtain complete face contour.The experimental results show that under different lighting conditions,the highest face contour recognition accuracy of this method is 96%,so this method has high recognition accuracy.

关 键 词：光照变化 深度学习 人脸轮廓识别 高斯统计 

分 类 号：TN01[电子电信—物理电子学]