计量学报2025，Vol.46Issue(11)：1612-1621,10.DOI:10.3969/j.issn.1000-1158.2025.11.09
基于低成本三维扫描仪的人脸深度图识别方法
Face Depth Map Recognition Method Based on Low-cost 3D Scanners
摘要
Abstract
A face depth map recognition(FDMR)method based on low-cost 3D scanners is proposed,aiming to enhance the robust feature expression ability of low-quality face depth maps to improve recognition accuracy,and to enhance the real-time performance of the algorithm to better meet practical application needs.Firstly,addressing the limitation of extracting features from single-depth data,low-quality deep face images are constructed as contrastive image pairs and fed into the FDMR network.During feature extraction,a continuous normalizing flow module is introduced to reduce the noise impact from low-quality deep face data.Lastly,a contrastive loss function tailored for image pairs is proposed.By minimizing the KL divergence between the similarity score distribution of image pairs and the matching distribution of real labels,the computation is simplified,thereby further enhancing training efficiency of the model.The proposed method demonstrates higher real-time performance compared to other current state-of-the-art methods,achieving a frame rate of 183 frames per second for image processing,whereas the current best method achieves 125 frames per second.Experiments conducted on the Lock3DFace and Extended-Multi-Dim datasets show that the proposed method improves the average recognition rate by 0.23% and 0.67%,respectively,compared to the current best method.关键词
低成本三维扫描仪/低质量人脸深度图像/对比图像对/特征去噪/对比损失函数/高实时性Key words
low-cost 3D scanner/low-quality face depth map/contrasting image pair/feature denoising/contrastive loss function/high real-time performance分类
通用工业技术引用本文复制引用
ZHANG Nannan,SANG Gaoli..基于低成本三维扫描仪的人脸深度图识别方法[J].计量学报,2025,46(11):1612-1621,10.基金项目
浙江省教育厅科研项目(Y202249424) （Y202249424）
嘉兴市公益性研究计划项目(2024AD10046) （2024AD10046）
