《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (5): 1645-1657.DOI: 10.11772/j.issn.1001-9081.2024050568

• 多媒体计算与计算机仿真 • 上一篇    

面向人脸识别的多模态研究方法综述

杨雅莉1, 黎英1,2,3(), 章育涛1, 宋佩华1,2,3   

  1. 1.南宁师范大学 物流管理与工程学院,南宁 530100
    2.广西高校智慧物流技术重点实验室(南宁师范大学),南宁 530100
    3.广西人机交互与智能决策重点实验室(南宁师范大学),南宁 530100
  • 收稿日期:2024-05-09 修回日期:2024-07-10 接受日期:2024-07-31 发布日期:2024-08-23 出版日期:2025-05-10
  • 通讯作者: 黎英
  • 作者简介:杨雅莉(1998—),女,河南信阳人,硕士研究生,主要研究方向:人脸识别、计算机视觉
    黎英(1973—),女,广西宁明人,副教授,博士,主要研究方向:医学影像处理、深度学习
    章育涛(1998—),男,浙江温州人,硕士研究生,主要研究方向:图像分析、深度学习
    宋佩华(1982—),男,江西抚州人,副教授,博士,主要研究方向:计算机图形学、深度学习。
  • 基金资助:
    国家自然科学基金资助项目(62062051);广西研究生教育创新计划项目(JGY2023222)

Review of multi-modal research methods for face recognition

Yali YANG1, Ying LI1,2,3(), Yutao ZHANG1, Peihua SONG1,2,3   

  1. 1.School of Logistics Management and Engineering,Nanning Normal University,Nanning Guangxi 530100,China
    2.Guangxi Key Laboratory of Intelligent Logistics Technology (Nanning Normal University),Nanning Guangxi 530100,China
    3.Guangxi Key Lab of Human-machine Interaction and Intelligent Decision (Nanning Normal University),Nanning Guangxi 530100,China
  • Received:2024-05-09 Revised:2024-07-10 Accepted:2024-07-31 Online:2024-08-23 Published:2025-05-10
  • Contact: Ying LI
  • About author:YANG Yali, born in 1998, M. S. candidate. Her research interests include face recognition, computer vision.
    LI Ying, born in 1973, Ph. D., associate professor. Her research interests include medical image processing, deep learning.
    ZHANG Yutao, born in 1998, M. S. candidate. His research interests include image analysis, deep learning.
    SONG Peihua, born in 1982, Ph. D., associate professor. His research interests include computer graphics, deep learning.
  • Supported by:
    National Natural Science Foundation of China(62062051);Innovation Program of Guangxi Graduate Education(JGY2023222)

摘要:

多模态人脸识别技术能充分利用人脸特征或其他生物特征提高识别的鲁棒性和安全性,具有广泛的实际应用价值。由于目前的多模态人脸识别研究存在模态差距和模态信息难以高效融合等问题,因此根据多种信息模态和应用目的对现有的多模态人脸识别方法进行分类综述,以梳理研究中存在的问题,并探讨未来的发展方向。首先,将基于多源信息融合的多模态人脸识别研究按照数据处理的不同阶段分为传感器级、特征级、评分级和决策级,并归纳现有方法的优势、局限性和适用场景;其次,将信息增强多模态人脸识别研究按照被增强模态的不同分为2D-3D信息增强和3D-2D信息增强,并总结现有方法的优缺点;再次,归纳总结基于其他生物特征和面向反欺诈的多模态人脸识别方法,并简要介绍常用的多模态人脸识别数据集相关信息;最后,给出多模态人脸识别研究中存在的一些严峻挑战,并展望未来的研究方向。

关键词: 多模态人脸识别, 特征融合, 信息增强, 生物特征, 反欺诈

Abstract:

Multi-modal face recognition technology can fully utilize face features and other biometric features to enhance recognition robustness and security, and has broad practical application value. Current research on multi-modal face recognition has problems such as modal disparity and inefficient modal fusion. Therefore, based on multiple information modalities and application purposes, the existing multi-modal face recognition methods were classified and reviewed to sort out the problems in research and explore future development directions. Firstly, the multi-modal face recognition research based on multi-source information fusion was divided into sensor-level, feature-level, scoring-level, and decision-level ones according to different stages of data processing, and advantages, limitations, and applicable scenarios of the existing methods were summarized. Secondly, the research on information-enhanced multi-modal face recognition was categorized into 2D-3D and 3D-2D information enhancement ones according to different enhanced modalities, and advantages and disadvantages of the existing methods were summed up. Thirdly, multi-modal face recognition methods based on other biometric features and for anti-spoofing were summarized, and the relevant information of commonly used multi-modal face recognition datasets were introduced briefly. Finally, key challenges and future development directions were given and prospected.

Key words: multi-modal face recognition, feature fusion, information enhancement, biometric feature, anti-spoofing

中图分类号: