Constrained Symmetric Non-Negative Matrix Factorization Guided by Pairwise Constraint Propagation for Image Clustering

To address the clustering of high-dimensional data, this paper proposes a novel semi-supervised clustering method named Constrained Symmetric Non-negative Matrix Factorization guided by Pairwise Constraint Propagation (PCSNMF). PCSNMF utilizes two types of supervision (pairwise constraints and label information) within the SNMF framework, leveraging the pairwise propagation algorithm and one-hot encoding. An optimization algorithm for PCSNMF is proposed, along with convergence and complexity analysis. Extensive experiments on four real-world image datasets demonstrate that PCSNMF outperforms existing state-of-the-art methods based on NMF and SNMF. Furthermore, runtime analysis and convergence experiments confirm the algorithm’s computational efficiency and fast convergence.