摘要
以基于低秩稀疏表示的子空间学习为研究对象,对近几年的相关研究工作进行了归纳总结。首先,阐述了子空间学习及低秩稀疏表示的概念;然后,根据迭代更新的方法,将基于低秩稀疏表示的子空间学习分为基于矩阵分解的子空间学习和基于谱聚类的子空间学习两大类;其次,对它们各自算法的核心思想进行了详细介绍,并对这些算法的优缺点进行了对比分析;最后,介绍了基于低秩稀疏表示的子空间学习在人脸识别、语音情感识别和运动分割这些领域的应用,同时指出了该研究中存在的挑战及未来研究方向。
The subspace learning based on low-rank representation and sparse representation is the research object,and the related research work in recent years were summarized.Firstly,the concepts of subspace learning and low rank sparse representation were explained. Secondly,according to the iterative update method,subspace learning based on low rank sparse representation was divided into two categories:subspace learning based on matrix decomposition and subspace learning based on spectral clustering.Then,the core ideas of their respective algorithms were introduced in detail and the advantages and disadvantages of these algorithms were compared and analyzed.Finally,the applications of this technology in face recognition,speech emotion recognition and motion segmentation were introduced.At the same time,the challenges and future research directions of subspace learning were presented.
作者
武继刚
陈招红
孟敏
谢敬豪
WU Jigang;CHEN Zhaohong;MENG Min;XIE Jinghao(School of Computer Science and Technology,Guangdong University of Technology,Guangdong 510006,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2021年第2期1-19,共19页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(62072118)
广东省自然科学基金重点项目(2018B030311007)
广东省科技计划重点领域研发计划资助项目(2019B010121001)。
关键词
子空间学习
子空间聚类
维数约简
低秩表示
稀疏表示
subspace learning
subspace clustering
dimensionality reduction
low-rank representation
sparse representation
