Comprehensive Low-Rank Tensor and Sparse Representation Recovery for Incomplete Multi-View Clustering

Incomplete multi-view clustering (IMVC) has become a key research area due to the missing views in real-world datasets. Traditional methods usually do not fully consider the correlation of data between different views or only focus on the low-rank structures in the intra-view graphs while ignoring the low-rank structures embedded in inter-view graphs. In order to solve these problems, we introduce a Comprehensive Low-rank Tensor and Sparse Representation Recovery for Incomplete Multi-View Clustering (CLTLSRR) approach. This method reformulates the IMVC problem as a joint incomplete similarity graph learning and comprehensive low-rank complete tensor representation recovery problem. It recovers tensor graphs by exploiting the low-rank structure in both inter-view (horizontal and lateral slices) and intra-view (frontal slices) graphs. It learns intra-view similarity graphs and constructs third-order tensor by stacking these graphs to reveal inter-view correlations. To mitigate the adverse effects of missing views and data noise, CLTLSRR decomposes the tensor into a sparse tensor and a low-rank intrinsic tensor. The proposed method proves its feasibility and effectiveness on several datasets.