Rotation Invariant Spatio-Spectral Total Variation for Hyperspectral Image Denoising

We propose a novel regularization function, named Rotation-Invariant Spatio-Spectral Total Variation (RISSTV), for hyperspectral (HS) image denoising. Spatio-Spectral Total Variation (SSTV) and its extended methods, defined using the second-order spatio-spectral differences, are known as popular regularization approaches that capture the HS image-specific spatio-spectral piecewise smoothness for effectively removing noise from HS images. However, these methods lack rotation-invariance and tend to corrupt round structures and oblique edges on HS images. To address this issue, we assign rotation-invariance to TV and establish RISSTV, which can accurately recover the detailed structures in HS images. Furthermore, we formulate the HS image denoising problem as a convex optimization problem that includes RISSTV and develop an efficient algorithm based on a preconditioned primal-dual splitting method to solve this problem efficiently. Finally, we demonstrate the effectiveness of RISSTV compared with existing HS image regularization models through mixed noise removal experiments.