Robust low rank tensor multi-view clustering

Abstract

Multi-view Spectral Clustering (MVSC) is a hot research direction in computer vision and machine learning. In recent years, scholars have proposed many MVSC methods based on tensor low rank representation. However, most of them are more suitable for processing noiseless data, but not ideal for noisy data. Inspired by the noise representation idea of hyperspectral noise images, this paper proposes a robust low rank tensor MVSC method for Gaussian and salt and pepper noise data based on MVSC-TLRN method. Similar to MVSC-TLRN method, the proposed method represents the multi-view clustering problem of noise data as a low rank tensor learning problem, which is solved by inexact augmented Lagrangian method. The experimental results on five image datasets and two document datasets show that the proposed method is much better than the existing methods.