Multi-view Subspace Clustering with Low-rank Kernel and High-order Similarity Learning
Yueyao Li, Bin Wu, Yanying Mei, Jiao Wang · 2023
Subspace clustering methods are effective ways to achieve clustering of high-dimensional data. The goal of current subspace clustering is to uncover potential structures hidden from diverse views, which can only be clustered in linear space. However, data points are typically collected from non-linear environments in real-world applications, leading to unsatisfactory clustering results. Such methods primarily focus on learning the direct similarity of the samples, ignoring the latent high-order correlations across views. As a result, it is challenging to visualize a distinct cluster structure. To this end, we offer a novel multi-view subspace clustering based on low-rank kernel and high-order similarity learning (LHMSC) in this paper. To mine the data worldwide structure in kernel space, the underlying consensus structure between various data points, and the inherent information of each view, it combines low-rank kernel learning and high-order similarity (HOS) learning into a single framework. Extensive tests on five real datasets reveal that our LHMSC method frequently outperforms state-of-the-art techniques.