遮挡人脸识别算法改进方法综述  

A Survey of Improved Methods of Occlusion Face Recognition Algorithms

作　　者：朱孟刚 郑广海[1] 

机构地区：[1]大连交通大学软件学院,辽宁 大连

出　　处：《计算机科学与应用》2022年第6期1569-1579,共11页Computer Science and Application

摘　　要：随着硬件水平的快速发展,人脸识别也有了大规模应用,通过人脸特征进行身份认证一直以来都是非常活跃的研究主题。而在实际环境下,许多外部因素都会对人脸识别任务产生不同程度的不良影响,例如:光线照射、表情变化、佩戴饰品遮挡等,而针对人脸部分的遮挡识别问题一直没有被完美解决。目前主流的遮挡人脸识别算法主要从以下四个角度进行研究:基于稀疏表示的遮挡人脸识别算法、基于主成分分析的遮挡人脸识别算法、基于生成对抗网络的遮挡人脸识别算法、基于卷积神经网络的遮挡人脸识别算法。本文将从上述4类算法进行了汇总,并介绍了各种算法的基本框架和设计原理,还分析了现阶段所面临的技术问题以及未来改进的方向。With the rapid development of hardware level, face recognition has also been widely used. Identity authentication through face features has always been a very active research topic. In the actual environment, many external factors will have different degrees of adverse effects on the face recognition task, such as light irradiation, expression change, wearing jewelry occlusion, etc. However, the occlusion recognition problem for the face has not been perfectly solved. At present, the main-stream occlusion face recognition algorithms are mainly studied from the following four perspectives: occlusion face recognition algorithm based on sparse representation, occlusion face recognition algorithm based on principal component analysis, occlusion face recognition algorithm based on generation countermeasure network and occlusion face recognition algorithm based on convolution neural network. This paper summarizes the above four kinds of algorithms, introduces the basic framework and design principle of various algorithms, and analyzes the technical problems faced at this stage and the direction of improvement in the future.

关 键 词：部分遮挡 人脸识别 生成对抗网络 稀疏表示 卷积神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]