Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (24): 27-38.DOI: 10.3778/j.issn.1002-8331.2107-0043

• Research Hotspots and Reviews • Previous Articles     Next Articles

Advances of Age Progression Based Cross-Age Face Recognition

LI Junyao, LI Zhihui, XIE Lanchi, HOU Xinyu, YE Dong   

  1. 1.Department of Criminal Science and Technology, Jiangsu Police Institute, Nanjing 210000, China
    2.Institute of Forensic Science, the Ministry of Public Security, Beijing 100032, China
  • Online:2021-12-15 Published:2021-12-13

基于老化模型的跨年龄人脸识别研究进展

李俊瑶,黎智辉,谢兰迟,侯欣雨,叶东   

  1. 1.江苏警官学院 刑事科学技术系,南京 210000
    2.公安部物证鉴定中心,北京 100032

Abstract:

Cross-age face recognition is one of the most difficult problems in face recognition at present. Face features will change with age, resulting in a decrease in recognition accuracy. The face recognition based on the generated aging images obtained from aging models is a popular solution for this problem. With the widespread application of computer technology and deep learning, the authenticity, aging effect, and algorithm efficiency of face aging have been significantly improved. This paper systematically reviews the current research status of cross-age face recognition based on aging models. The aging methods are sorted out in detail, the method evolution of aging models and the advantages and disadvantages of various methods are systematically introduced, and the existing model evaluation methods are summarized. In addition, the existing datasets that can be used for cross-age face recognition are introduced in detail, and a comparative analysis is made in terms of data volume, age span, age accuracy, and use of the data set. For the purpose of practical applications, the problems to be solved in cross-age face recognition based on the aging model are analyzed and discussed. Moreover, the future research directions are predicted and prospected.

Key words: face aging, cross-age, deep learning, face recognition

摘要:

跨年龄人脸识别是目前人脸识别中的一大难点问题,人脸特征会随着年龄的增长发生变化,导致识别准确率降低,利用老化模型生成老化图像后进行人脸识别为该问题提供了一种解决方案。随着计算机技术和深度学习的广泛应用,人脸老化的真实性、老化效果、算法效率都得到了明显的提升,系统综述了基于老化模型的跨年龄人脸识别的研究现状,对人脸老化方法进行了详细地梳理,系统介绍了老化模型的方法演变和各类方法的优缺点,并对现有的模型评价方法进行了总结归纳。对现有的可用于跨年龄人脸识别的数据集进行了详细介绍,从数据量、年龄跨度、年龄准确性、数据集使用情况等方面进行了对比分析。结合实际应用对基于老化模型的跨年龄人脸识别中待解决的问题进行了分析和讨论,并对未来研究方向做出预测和展望。

关键词: 人脸老化, 跨年龄, 深度学习, 人脸识别