Multi-View Subspace Clustering via Anchor Graph Factorization

In recent years, multi-view subspace clustering has become a research hotspot due to its excellent performance in handling high-dimensional data. Among them, anchor-based methods play an important role on the grounds of the scalability, but face the limitation of disjoint optimizations. To address this shortcoming, we propose Multi-view Subspace Clustering via Anchor Graph Factorization (MAGF), which can complete clustering without any pre- or post-processing. Specifically, this work enables anchor graph backtracking based on the indicator matrix, thus incorporating optimization into the unified framework, rather than independent stages. In this way, our method achieves strong clustering performance and high efficiency compared with other state-of-the-art methods, which can be verified based on the extensive experiments on five benchmark datasets.