Lv Ziyu, Gao Quanxue, Zhang Xiangdong, Li Qin, Yang Ming
IEEE Trans Image Process. 2022;31:4790-4802. doi: 10.1109/TIP.2022.3187562. Epub 2022 Jul 15.
In this article, we present a novel general framework for incomplete multi-view clustering by integrating graph learning and spectral clustering. In our model, a tensor low-rank constraint are introduced to learn a stable low-dimensional representation, which encodes the complementary information and takes into account the cluster structure between different views. A corresponding algorithm associated with augmented Lagrangian multipliers is established. In particular, tensor Schatten p -norm is used as a tighter approximation to the tensor rank function. Besides, both consistency and specificity are jointly exploited for subspace representation learning. Extensive experiments on benchmark datasets demonstrate that our model outperforms several baseline methods in incomplete multi-view clustering.
在本文中，我们提出了一种通过整合图学习和谱聚类来进行不完全多视图聚类的新型通用框架。在我们的模型中，引入了张量低秩约束来学习稳定的低维表示，该表示编码了互补信息并考虑了不同视图之间的聚类结构。建立了一种与增广拉格朗日乘子相关的相应算法。特别地，张量Schatten p -范数被用作张量秩函数的更紧密近似。此外，在子空间表示学习中同时利用了一致性和特异性。在基准数据集上进行的大量实验表明，我们的模型在不完全多视图聚类方面优于几种基线方法。
