基于低秩张量表示的多视图子空间聚类

Multi-View Subspace Clustering Based on Low-Rank Tensor Representation

近年来,多视图子空间聚类是一个热点话题,基于低秩张量的方法受到广泛关注。为了更好地挖掘不同视图间的高阶关联性,本文采用最新基于t-SVD的张量核范数,使用系数矩阵的核范数和Frobenius范数作为正则项。在PIE、ORL、MSRA和MNIST四个数据集上与流行的子空间聚类算法的对比试验表明,本文提出的算法是一个有效的方法。

In recent years, multi-view subspace clustering has been a hot topic, and methods based on low-rank tensors have received widespread attention. In order to better mine the high-order cor-relation between different views, this paper adopts the latest tensor kernel norm based on t-SVD, using the kernel norm and Frobenius norm of the coefficient matrix as regularization terms. Comparative experiments with popular subspace clustering algorithms on four data sets: PIE, ORL, MSRA and MNIST show that the algorithm proposed in this article is an effective method.