Hypergraph regularization-based anchor learning for multi-view clustering

Current anchor graph-based multi-view clustering methods can effectively address the problem of high computational cost for clustering large-scale multimedia data. However, they have the following shortcomings: (1) The relationships between anchor points are not adequately considered. (2) The correlations between the consistent anchor graph and diverse anchor graphs are ignored. To handle these issues, we propose a novel multi-view clustering method named Hypergraph Regularization-Based Anchor Learning (HRFAL). Specifically, we first process the original data to obtain a consistent anchor graph and diverse anchor graphs, which can explore more comprehensive consistent and complementary information. Meanwhile, the hyper-Laplacian regularization is applied to the anchor points to explore the higher-order relationships between the anchor points, thus enabling the generation of high-quality anchor graphs. Furthermore, the orthogonal diversity constraints are imposed on the consistent and diverse anchor graphs to enhance the distinction between the consistent and diverse components, resulting in better exploitation of consistent and complementary information. Finally, the Schatten $p$-norm constraint is implemented on the consistent anchor graph to maintain its low-rank structure, thus obtaining more robust consistent information. Experimental results on eight multi-view datasets show that HRFAL exhibits superior performance in terms of accuracy and speed.