Global and local structure preserving nonnegative subspace clustering

Hongjie Jia, Dongxia Zhu, Longxia Huang, Qirong Mao, Liangjun Wang, Heping Song

Abstract: Highlights•A novel global and local structure preserving nonnegative subspace clustering model is designed.•Nonnegative subspace clustering can directly obtain the cluster indicators and allocate cluster members.•Efficient multiplicative updating rules are developed by nonnegative Lagrangian relaxation to optimize the model.•The data similarities and cluster indicators are learned simultaneously and can promote each other through iterative optimization.•The effectiveness of the model is demonstrated by theoretical analysis and comprehensive experiments.