自加权的缩放单形表示子空间聚类算法

Subspace clustering can assign high-dimensional data to different low-dimensional subspaces, which has extensive applications. The majority of subspace clustering techniques usually make the assumption that each variable in high-dimensional data has an equal impact on the clustering process. However, this assumption is not suitable for practical applications. To address the above issue, this paper proposes a self-weighted scaled simplex representation subspace clustering method. The self-expressive method is used to reconstruct the weighted data after each variable has been given an appropriate weight based on differences in relevance. In addition, a sparsity regularization term is utilized to control the sparsity of weights. Simultaneously, the scaled simplex representation is introduced to obtain a more reliable coefficient matrix. The enhanced Lagrangian multipliers approach is used to optimize all of these phases and combine them into a single framework. Experimental results on real-world datasets demonstrate that the proposed algorithm has better clustering performance than existing clustering methods.