Low Rank and Semi-Nonnegative Tensor Factorization for Hyperspectral Imagery Denoising

Aihong Yuan, Junchao Yu, Xiaoyi Zuo, Kaixin Gao, Mengbo You

Abstract: Hyperspectral image denoising can play an important role in addressing the issue of preprocessing massive high-dimensional hyperspectral data for subsequent object detection or classification. This paper presents a novel low-rank and semi-nonnegative tensor factorization method for HSI denoising. The framelet regularization is introduced to constrain the reduced-dimensionality factor, rather than directly regularizing HSI itself in the low-rank and non-negative tensor factorization model. Thus, our method preserves the details and geometric features of restored HSI in the spatial domain and demand much less calculation and computer memory. Extensive experimental results show that our method is superior to other existing methods for HSI denoising in simulated benchmark datasets. Our source code is available at: https://github.com/misteru/LRSNTF.