Robust subspace clustering based on inter-cluster correlation reduction by low rank representation

Abstract

Subspace clustering refers to clustering data points into their respective subspaces and finding a low-dimensional structure to fit each group of points. In subspace clustering, the inter-cluster correlation of data which is caused by noise such as illumination and background affects the performance of subspace clustering algorithms. To solve this problem, a new approach is proposed to detect the unusual data with strong inter-cluster correlation based on the representation matrix obtained from low-rank representation (LRR). Then a low-rank model was established by reducing the unusual part of data and subspace clustering is performed. In addition, in order to apply subspace clustering algorithm on unaligned data, the preprocessing is required to make the data aligned, and the preprocessed data is used in experiments. Experimental results on the face dataset and texture dataset show the efficiency of the proposed method. The experiment also indicates that with the signal ratio increasing, the inter-cluster correlation is becoming more obvious.