Feature concatenation multi-view subspace clustering

Abstract

Multi-view clustering aims to achieve more promising clustering results than single-view clustering by exploring the multi-view information. Since statistic properties of different views are diverse, even incompatible, few approaches implement multi-view clustering based on the concatenated features directly. However, feature concatenation is a natural way to combine multiple views. To this end, this paper proposes a novel multi-view subspace clustering approach dubbed Feature Concatenation Multi-view Subspace Clustering (FCMSC). Specifically, by exploring the consensus information, multi-view data are concatenated into a joint representation firstly, then, l2,1-norm is integrated into the objective function to deal with the sample-specific and cluster-specific corruptions of multiple views for benefiting the clustering performance. Furthermore, by introducing graph Laplacians of multiple views, a graph regularized FCMSC is also introduced to explore both the consensus information and complementary information for clustering. It is noteworthy that the obtained coefficient matrix is not derived by directly applying the Low-Rank Representation (LRR) to the joint view representation simply. Finally, an effective algorithm based on the Augmented Lagrangian Multiplier (ALM) is designed to optimized the objective functions. Comprehensive experiments on six real world datasets illustrate the superiority of the proposed methods over several state-of-the-art approaches for multi-view clustering.

Non-negative sparse Laplacian regularized latent multi-view subspace clustering

“…-RMSC 5 : This method first clusters each view, and then uses the shared low-rank transition probability matrix to integrate them to get better clustering results. -FCMSC 27 : The algorithm deals with the problem that multiple views have different statistic properties Input: Multi-view matrices:…”

Kernel‐based low‐rank tensorized multiview spectral clustering

“…Zhang et al 37 explored high‐order correlations of multiview data by regarding subspace representation matrices as a low‐rank tensor. Zheng et al 38 concatenated multiview data into a joint representation and explored both the consensus information and complementary information of multiple views. Kang et al 39 fused multiview information in partition level and integrated graph learning from each view, the generation of basic partitions, and the merging of consensus partition.…”

Multiview spectral clustering via complementary information

“…Most existing multiview clustering methods can be classified into three major categories, namely, multiview spectral clustering methods, 8,9,[11][12][13] multiview subspace clustering methods, 7,[19][20][21][22][23] and multiview nonnegative matrix factorization clustering methods. [24][25][26][27][28][29] Although the three categories of methods are based on diverse theories, they have the same main idea, in other words, all these methods are designed to combine information from multiview data (or representations) into a common representation, pursuing the consensus information among all views.…”