计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (22): 14-24.DOI: 10.3778/j.issn.1002-8331.1906-0005

• 热点与综述 • 上一篇    下一篇

低分辨人脸识别综述

张凯兵,郑冬冬,景军锋   

  1. 西安工程大学 电子信息学院,西安 710048
  • 出版日期:2019-11-15 发布日期:2019-11-13

Survey of Low-Resolution Face Recognition

ZHANG Kaibing, ZHENG Dongdong, JING Junfeng   

  1. School of Electronics and Information, Xi’an Polytechnic University, Xi’an 710048, China
  • Online:2019-11-15 Published:2019-11-13

摘要: 全面综述了低分辨(Low-Resolution,LR)人脸识别技术的研究进展,并对相关亟需解决的关键问题进行了讨论。对LR人脸识别系统的概念、待解决问题、系统结构、已有不同识别方法进行了分类阐述。根据高、低分辨率人脸图像空间特征维度的不匹配问题,分别对基于重构超分辨(Super-Resolution,SR)图像和基于公共特征子空间两类LR人脸识别方法进行了详细介绍。对每类方法按照不同的实现过程,进一步划分为三种不同的类型分别介绍,并对每类方法模型的主要思想和核心问题进行了分析和讨论。简单介绍了九种标准人脸数据库,从识别性能、平均运行时间和多人脸库实验结果比较等方面对每类代表性方法进行了分析。对LR人脸识别技术在未来发展中需要解决的难点问题给予了展望。

关键词: 低分辨人脸识别, 图像超分辨, 耦合映射, 字典学习, 稀疏表示, 神经网络

Abstract: This paper comprehensively surveys the research progress of Low-Resolution(LR) face recognition methods, and discusses the unsolved issues. Firstly, the fundamental concepts, the unsolved issues, the system’s structure, and the existing LR face recognition methods are presented, respectively. Secondly, in terms of the characteristics of mismatching between the High-Resolution(HR) and LR face images, two major categories of LR face recognition methods based on Super-Resolution(SR) image and common feature subspace are introduced in detail. According to different implementation processes, each category method is further divided into three different types, and the main ideas and core issues of each kind of method are analyzed and discussed. Then nine kinds of standard face databases are briefly introduced, and the experimental results of each representative method on the introduced face databases are analyzed from the aspects of recognition performance, average running time, and experimental comparisons performed on several face databases. Finally, the difficulties to be solved in the future development of LR face recognition technology are prospected.

Key words: low-resolution face recognition, image super-resolution, coupled mapping, dictionary learning, sparse representation, neural network