基于优化形变统计模型的3D人脸识别  

3D Face Recognition Based on Optimal Deformation Statistical Model

作　　者：蔡梦园 袁三男 CAI Mengyuan;YUAN Sannan(College of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海201306

出　　处：《电子器件》2025年第1期116-122,共7页Chinese Journal of Electron Devices

摘　　要：为了解决大型3D人脸数据集缺乏导致的人脸识别精度不高的问题,提出一种基于优化形变统计模型(GPSM)的3D人脸识别方法。首先,在参考数据集中,应用GPSM得到3D人脸的形状GPSM模型、3D人脸表情GPSM模型和3D人脸的质地GPSM模型,将这三种GPSM模型进行权重线性叠加,生成大量的3D人脸,形成大型3D人脸数据集;接着,对GPSM合成的人脸进行预处理;然后,用预处理后的人脸来训练3D人脸识别网络Res-GLNet;最后,在公共数据集FRGCv2和Bosphorus上测试Res-GLNet性能。结果显示:所提方法分别获得了98.9%和98.67%的识别率,优于Pointnet,Pointnet++,Pointcnn等识别方法,由此证明所提方法使得识别精度得到较大的提升。In order to solve the problem of low accuracy of face recognition caused by the lack of large-scale 3D face datasets,a 3D face recognition method based on Gaussian statistical shape models(GPSM)is proposed.First,in the reference data set,GPSM is applied to obtain the 3D face shape GPSM model,the 3D face expression GPSM model and the 3D face texture GPSM model,and the three GPSM models are linearly combined to generate a large number of 3D faces,forming a large 3D face dataset.Next,the face synthesized by GPSM is preprocessed.Then,the preprocessed face is used to train the 3D face recognition network of Res-GLNet.Finally,the perform-ance of Res-GLNet is tested on the FRGCv2 and Bosphorus public datasets.The results show that the proposed method achieves 98.9%and 98.67%recognition rates respectively,which are better than Pointnet,Pointnet++,Pointcnn and other recognition methods,proving that the proposed method can greatly improve the recognition accuracy.

关 键 词：机器视觉 人脸合成 3D人脸识别 人脸点云 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]