Multi-view spectral clustering via robust local subspace learning

Because of the existence of multiple sources of datasets, multi-view clustering has a wide range of applications in data mining and pattern recognition. Multi-view could utilize complementary information that existed in multiple views to improve the performance of clustering. Recently, there have been multi-view clustering methods which improved the performance of clustering to some extent. However, they do not take local representation relationship into consideration and local representation relationship is a crucial technology in subspace learning. To solve this problem, we proposed a novel multi-view clustering algorithm via robust local representation. We consider that all the views are derived from a robust unified subspace and noisy. To get the robust similarity matrix we simultaneously take all the local reconstruction relationships into consideration and use L1-norm to guarantee the sparsity. We give an iteration solution for this problem and give the proof of correctness. We compare our method with a number of classical methods on real-world and synthetic datasets to show the efficacy of the proposed algorithm.