《计算机应用研究》|Application Research of Computers

基于可靠邻居与精确簇数的稀疏子空间聚类

Sparse subspace clustering based on reliable neighbors and exact cluster number

免费全文下载 (已被下载 次)  
获取PDF全文
作者 郑毅,马盈仓,杨小飞
机构 西安工程大学 理学院,西安 710600
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2021)01-015-0075-08
DOI 10.19734/j.issn.1001-3695.2019.10.0604
摘要 为了获得更加可靠的相似矩阵,并使其含有精确的连通分支数量,提出了一种新的稀疏子空间聚类算法。该算法利用K近邻思想从局部寻找可靠邻居,在距离度量方面,选用测地线距离进行计算,考虑了数据在高维空间分布的几何结构,使得数据的邻居关系更加合理。同时,利用Ky Fan定理,通过参数的自适应调节,使得相似矩阵包含精确的连通分支数量。此外,该算法打破了常规的两步走模式,同时进行相似矩阵的学习和谱聚类过程,将数据相似性度和分割进行了紧密的联系,进一步加强了对数据结构信息的挖掘和利用。在人造数据集、图像数据集以及真实数据集进行了实验,实验结果表明该算法是有效的。
关键词 K近邻; 测地线距离; 子空间聚类; 连通分支数量; 相似矩阵
基金项目 国家自然科学基金资助项目(11501435)
陕西省教育厅科研计划项目(18JS042)
陕西省重点研发计划项目(2018KW-021)
本文URL http://www.arocmag.com/article/01-2021-01-015.html
英文标题 Sparse subspace clustering based on reliable neighbors and exact cluster number
作者英文名 Zheng Yi, Ma Yingcang, Yang Xiaofei
机构英文名 School of Science,Xi'an Polytechnic University,Xi'an 710600,China
英文摘要 In order to obtain a more reliable similarity matrix and make it contain the exact number of connected branches, this paper proposed a new sparse subspace clustering algorithm. The algorithm used the K nearest neighbor idea to find reliable neighbors from the local. In the aspect of distance metric, the algorithm selected the geodesic distance for calculation, and considered the geometric structure of the data in the high-dimensional space, which made the neighbor relationship of the data more reasonable. At the same time, using the Ky Fan theorem, the adaptive matrix adjusted the parameters so that the similar matrix contained the exact number of connected branches. In addition, the algorithm broke the conventional two-step mode, and simultaneously performed similar matrix learning and spectral clustering process, which closely linked data similarity and segmentation, further strengthening the mining and utilization of data structure information. The results of the experiments on artificial datasets, image datasets and real datasets show that the algorithm is effective.
英文关键词 K nearest neighbor; geodesic distance; subspace clustering; number of connected branches; similarity matrix
参考文献 查看稿件参考文献
 
收稿日期 2019/10/31
修回日期 2019/12/6
页码 75-82
中图分类号 TP181
文献标志码 A