基于子空间融合的多视图聚类算法

Multi-view Clustering Algorithm Based on Subspace Fusion

提出一种新颖的多视图子空间聚类算法,不再对包含各种噪声以及冗余信息的原始数据进行特征融合,而是通过对不同视图的低维子空间表示进行融合,得到一个公共的低维子空间表示。将这个子空间表示作为相似度矩阵进行谱聚类,以得到更优的聚类效果。在3个广泛使用的多视图基准数据集上进行了实验,实验结果证明了所提出算法的有效性。

A novel algorithm of multi-view subspace clustering was proposed,which no longer fused the features of original data containing various noises and redundant information.Instead,the low-dimensional subspace representation of different views was fused and a common low-dimensional subspace representation was obtained.Then the subspace representation as a similarity matrix for spectral clustering was used to obtain a better clustering effect.Experimental results on three widely used multi-view benchmark data sets showed the effectiveness of the proposed algorithm.