Multi-view kernel subspace clustering with adaptive information completion and fusion for unsupervised systems
Abstract Subspace clustering methods are increasingly favored in engineering applications because of their unsupervised nature. However, their performance in processing multi-view nonlinear data in unsupervised systems is often affected by the following three factors: (1) how to use data with valid...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-08-01
|
| Series: | Journal of King Saud University: Computer and Information Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s44443-025-00126-y |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Subspace clustering methods are increasingly favored in engineering applications because of their unsupervised nature. However, their performance in processing multi-view nonlinear data in unsupervised systems is often affected by the following three factors: (1) how to use data with valid information from each view in an appropriate way to avoid introducing irrelevant information; (2) how to improve incomplete data information in each view; and (3) how to balance the contributions of different views to ensure that the most informative view receives adequate attention. In this research, we propose a multi-kernel RPCA with graph regulation to prevent irrelevant information from being introduced into the clustering process of target objects. We also design a multi-view data completion framework with enhanced low-rank constraint to improve the quality of incomplete data in each view. Furthermore, we integrate these components into a unified framework and emphasize the effective synergy and complementary integration of information across different views. A large number of experiments in unsupervised systems, including multi-view data clustering, color video motion segmentation, invisible light video motion segmentation and color image segmentation, demonstrate the feasibility of our proposed method. |
|---|---|
| ISSN: | 1319-1578 2213-1248 |