Abstract:

A new subspace clustering model Structured Weighted Correlation Adaptive Subspace Clustering（SWCASC） is proposed to solve the problem that the structural sparse subspace clustering cannot well grasp the consistency of data similarity. In this model, the correlation coefficient of data points is introduced to impose an explicit penalty on the representation coefficient, and the subspace structure norm is introduced for the dependency relation between segmentation and similarity. The new model makes the data category labels consistent, the similarity matrix sparse and consistent, and self-adaptive. The sparsity of similarity matrix is conducive to separating data from different subspaces, while consistency is conducive to aggregating data from the same subspace. Extensive experiments show that the method has better classification performance and is superior to other methods.

Key words: subspace clustering, correlation coefficient, data similarity, consistency

摘要：

针对结构稀疏子空间聚类不能很好地把握数据相似度一致性的问题，提出一种新的子空间聚类优化模型；结构加权相关自适应子空间聚类（Structured Weighted Correlation Adaptive Subspace Clustering，SWCASC）模型。该模型引入数据点的相关性对表示系数施加显式惩罚，同时利用分割和相似度的依赖关系，引入子空间结构范数。该模型使得数据类别标签具有一致性，相似度矩阵具有稀疏性和一致性，并具有自适应性。相似度矩阵的稀疏性有利于将不同子空间的数据分离，而一致性有利于将同一子空间的数据聚集。实验结果表明，该模型获得了理想的聚类效果，并优于其他方法。

关键词: 子空间聚类, 相关系数, 数据相似度, 一致性