基于特征级联的联合平滑多视图子空间聚类算法

Jointly smoothed multi⁃view subspace clustering based on feature concatenation

近年来,多视图聚类问题受到国内外的广泛关注.联合平滑多视图子空间聚类算法通过视图共识分组效应,利用多个视图的局部结构来规范视图共性表示,取得了不错的聚类效果,但是该算法对于不一致性的探索仍然存在一定的局限性,限制了聚类性能的进一步提升.为了进一步挖掘多视图的不一致性,提出一种基于特征级联的联合平滑多视图子空间聚类算法,它不仅同时学习视图间的一致性与不一致性,增强视图的多样性,还将不一致性划分为特定于集群的不一致性与特定于样本的不一致性,通过核范数进一步与低秩表示相关联,并在此基础上使用交替方向最小化进行迭代.在四个公共数据集上与其他优秀算法进行了对比实验,证明了所提算法的优越性.

In recent years,the multi⁃view clustering problem has received widespread attention both domestically and internationally.The jointly smoothed multi⁃view clustering algorithm utilizes the view⁃consensus grouping effect and the local structure of multiple views to standardize the common representation of views,achieving impressive clustering results.However,this algorithm still has certain limitations in exploring inconsistency,which limits further improvement of the clustering performance.In order to further explore the inconsistency of multiple views,this paper proposes a jointly smoothed multi⁃view subspace clustering algorithm based on feature concatenation.It not only learns the consistency and inconsistency between views simultaneously to enhance view diversity,but also divides the whole inconsistency into cluster⁃specific and sample⁃specific corruptions.It is further associated with low⁃rank representations through kernel norm,and on such a basis of iterates using alternating direction minimization.Experiments conducted on four benchmark datasets have demonstrated the superiority of the proposed algorithm over other excellent algorithms.