Subspace Clustering Based on Latent Low Rank Representation with Schatten-p Norm

Abstract

Clustering analysis is one of the key technologies in the field of data mining. Among them, high-dimensional data clustering is the core and the most challenging task in clustering analysis. Subspace clustering is an effective clustering method for high-dimensional data. Subspace clustering with Latent Low-Rank Representation (LatLRR) is promising because it can solves the problem of Low-Rank Representation (LRR) of insufficient samples. However, the nuclear norm is usually used to approximate the rank in LatLRR since finding the low rank solution is NP-hard. In order to obtain a better low rank matrix and take into account the insufficiency of samples, this paper proposes a LatLRR model based on Schatten-p norm which introduces the Schatten-p norm to approximate the rank function and an Lp norm constraint error term to improve the robustness. Experimental results show that the algorithm can effectively improve the subspace clustering performance.