Specific and coupled double consistency multi-view subspace clustering with low-rank tensor learning

Abstract

Multi-view clustering (MVC) has gained widespread attention due to its ability to utilize different features from different views. However, the existing MVC methods fail to fully exploit the consistency across multiple views, leading to information loss. Additionally, the performance of the algorithms is not satisfactory due to the inherent noise in the data. To address the above-mentioned issues, this paper proposes the specific and coupled double consistency multi-view subspace clustering with low-rank tensor learning (SCDCMV) method. Specifically, firstly, we simultaneously incorporate the consistency and specificity of multiple views into self-expressive learning. However, the information within the consistency matrix has not been fully utilized, and there still exists some noise. Then, the obtained consistency matrix is once again integrated into self-expressive learning to obtain a new consistency matrix. Thirdly, we combine the two consistency matrices into a tensor and constrain it using tensor nuclear norm (TNN). Then, under the constraint of TNN, the two consistency matrices mutually reinforce each other, which helps fully utilize the consistency information and reduce the impact of noise, ultimately leading to better clustering results. Ultimately, these three steps constitute a framework that is tackled utilizing the augmented Lagrange multiplier method. The performance of SCDCMV has improved by 55.94 %. Experimental results on different datasets indicate that the SCDCMV algorithm outperforms state-of-the-art algorithms. In other words, these experimental results validate the importance of effectively utilizing consistent information from multiple views while reducing the impact of noise. The code is publicly available at https://github.com/TongWuahpu/SCDCMV.