Multi-View Subspace Clustering Via A Global Low-Rank Affinity Matrix

Subspace clustering is a technique which aims to find the underlying low-dimensional subspace in a high-dimensional data space. Since the multi-view data exists generally and it can effectively improve the performance of the learning task in real-world applications, multi-view subspace clustering has gained lots of attention in recent years. In this paper, to further improve the clustering performance of multi-view subspace clustering, we propose a novel subspace clustering method based on a global low-rank affinity matrix. In our method, we introduce a global affinity matrix, and use a sparse term to fit the difference between the global affinity matrix and local affinity matrices. Meanwhile, our method explores the global consistent information from different views and simultaneously guarantees the global affinity matrix for segmentation is low-rank. The objective function can be solved efficiently by the inexact augmented Lagrange multipliers (ALM) optimization method. Experiments results on two public real face datasets demonstrate that our method can improve the clustering performance against with the state-of-the-art methods.