基于核的低秩子空间聚类算法

Kernel-based Low-rank Representation Subspace Clustering Algorithm

基于稀疏表示和低秩表示的子空间聚类算法是目前的研究热点,但大多数子空间聚类方法只适用于线性子空间或仿射子空间。针对这一问题,研究了一种能处理非线性模型的核子空间聚类方法。提出学习一种低秩核映射,通过这种映射,特征空间中的映射数据不仅具有低秩性,而且具有自表达性,从而使得低维子空间结构在高维特征空间中得以呈现。通过运动分割和人脸图像聚类问题的实验,验证了方法的有效性。

Subspace clustering algorithms based on sparse representation or low-rank representation have be-come a current research hotspot,but most subspace clustering methods are only suitable for linear subspace or affine subspace.In order to solve this problem,this paper researches a kernel low-rank representation subspace clustering method(KLRR)that can deal with nonlinear model and propose a low-rank kernel mapping.Through this mapping,the mapping data in the feature space not only has low rank property,but also has self-expression property,so that the low dimensional subspace structures are presented in high dimensional feature space.Experiments on motion segmentation and face image clustering show the effectiveness of the proposed method.