Graph Convolutional Subspace Clustering: A Robust Subspace Clustering Framework for Hyperspectral Image

Hyperspectral image (HSI) clustering is a challenging task due to the high complexity of HSI data. Subspace has been proven be powerful for exploiting intrinsic relationship between data points. Despite impressive performance in clustering, traditional subspace methods often ignore inherent structural information among In this article, we revisit with graph convolution and present novel framework called convolutional (GCSC) robust clustering. Specifically, recasts self-expressiveness property into non-Euclidean domain, which results more embedding dictionary. We show that models are special forms our Euclidean On basis framework, further propose two by using Frobenius norm, namely efficient GCSC (EGCSC) kernel (EKGCSC). Each model globally optimal closed-form solution, making it easier implement, train, apply practice. Extensive experiments strongly evidence EGCSC EKGCSC dramatically outperform current on three popular sets consistently.