Double Constrained NMF for Partial Multi-View Clustering
Bin Qian, Xiaobo Shen, Yanyang Gu, Zhenmin Tang, Yuhua Ding · 2016
Many computer vision problems involve data with multiple views, where each view corresponds to a certain type of feature. To integrate information from multiple views in the unsupervised setting, multi-view clustering methods have been developed to cluster multiple views simultaneously. Most existing methods only consider the case that each example appears in all the views. However, data with partial views is often occurred in real applications. For example, several certain sensors sometimes have faults, then the data may not be captured completely, which will lead to the case that only partial views are available. In this paper, we propose a nonnegative matrix factorization (NMF) method for partial multi-view clustering, which incorporates the cluster similarity and manifold preserving constraints into an unified framework. The basic idea of the proposed method, named Double Constrained NMF (DCNMF), is to push clustering solutions of different views from the same examples towards a common membership matrice, and to maintain the latent geometric structure of the views simultaneously. Moreover, we develop an efficient optimization scheme for the proposed method. Experiments on several two-view datasets demonstrate the advantages of the proposed method on partial multi-view clustering tasks.