一种改进的子空间聚类方法

An improved subspace clusteringmethod

目的更好地揭示高维数据的子空间结构,提高子空间聚类性能。方法对系数矩阵施加Frobenius范数约束,并使其与稀疏矩阵充分接近,建立新的子空间表示模型,利用系数矩阵构造相似度矩阵,最后利用谱聚类算法得到聚类结果。结果与结论新模型能得到类间稀疏和类内聚集的系数矩阵,提高了聚类性能,且能快速实现。

Purposes— To reveal the subspace structure of the high-dimensional data well, and to improve the performance of subspace clustering.Methods— The coefficient matrix is constrained with Frobenius norm, and is approximated to a sparse matrix. A new model is constructed for subspace representation, whose solution is used to determine an affinity matrix. The final clustering result is obtained by employing a spectral clustering algorithm.Results and Conclusion— The coefficient matrix obtained from the proposed model has such good properties as sparseness between clusters and grouping within clusters, which result in high clustering performance and fast computation.