作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (4): 56-61,67. doi: 10.19678/j.issn.1000-3428.0057476

• 人工智能与模式识别 • 上一篇    下一篇

结构约束的对称低秩表示子空间聚类算法

陶洋, 鲍灵浪, 胡昊   

  1. 重庆邮电大学 通信与信息工程学院, 重庆 400065
  • 收稿日期:2020-02-24 修回日期:2020-05-02 发布日期:2020-05-11
  • 作者简介:陶洋(1964-),男,教授、博士,主研方向为模式识别、异构网络、物联网;鲍灵浪、胡昊,硕士研究生。
  • 基金资助:
    重庆市自然科学基金(cstc2018jcyjAX0344)。

Structure-Constrained Symmetric Low-Rank Representation Algorithm for Subspace Clustering

TAO Yang, BAO Linglang, HU Hao   

  1. School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
  • Received:2020-02-24 Revised:2020-05-02 Published:2020-05-11

摘要: 通过子空间聚类可获得高维数据的潜在子空间结构,但现有算法不能同时揭示数据全局低秩结构和局部稀疏结构特性,致使聚类性能受限。提出一种结构约束的对称低秩表示算法用于子空间聚类。在目标函数中添加结构约束和对称约束来限制低秩表示解的结构,构造一个加权稀疏和对称低秩的亲和度图,在此基础上,结合谱聚类方法实现高效的子空间聚类。实验结果表明,该算法能够准确表示复杂子空间结构,其在Extended Yale B和Hopkins 155基准数据集上的平均聚类误差分别为1.37%和1.43%,聚类性能优于LRR、SSC、LRRSC等算法。

关键词: 低秩表示, 稀疏表示, 加权约束, 对称约束, 子空间聚类

Abstract: The potential subspace structure of high-dimensional data can be obtained by using subspace clustering,but the existing methods can not reveal the characteristics of global low-rank structure and local sparse structure of data at the same time,which limits the clustering performance.This paper proposes a Structure-Constrained Symmetric Low-Rank Representation(SCSLR) algorithm for subspace clustering.The structure constraint and symmetry constraint are introduced into the object function to limit the solution structure of low-rank representation,and a weighted sparse and symmetric low-rank affinity graph is constructed.On this basis,the spectrum clustering method is used to realize efficient subspace clustering.Experimental results show that the proposed algorithm can accurately represent the complex subspace structure.Its average clustering error on two benchmark datasets,Extended Yale B and Hopkins 155,is 1.37% and 1.43% respectively,and its clustering performance is better than that of Low-Rank Representation(LRR),Sparse Subspace Clustering(LSS),Structure-Constrained Symmetric LRR(LRRSC) and other algorithms.

Key words: Low-Rank Representation(LRR), sparse representation, weighed constraint, symmetric constraint, subspace clustering

中图分类号: