Spatio-Spectral Structure Tensor Total Variation for Hyperspectral Image Denoising and Destriping

This is a demo code of the proposed method in the following reference:

S. Takemoto, K. Naganuma, and S. Ono, ``Spatio-Spectral-Structure-Tensor-Total-Variation-for-Hyperspectral-Image-Denoising-and-Destriping,'' IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025.

Update history: 10, Aug. 2025: v1.0

For more information, see the following

• Project website: https://www.mdi.c.titech.ac.jp/publications/s3ttv
• Preprint paper: https://arxiv.org/abs/2404.03313

How to use

1. Setting parameters

• Choose the image (JasperRidge, PaviaUniversity, or Beltsville)
• Adjust the parameters
  • params.rho: parameter for the radii of the noise terms
  • params.blocksize: Block size of spatio-spectral structure tensor
  • params.stopcri: Stopping criterion
  • params.maxiter: Maximum number of iterations
  • params.disprate: Period to display intermediate results
• Set as use_GPU = 1 if you use GPU.
• Set as use_fast = 1 if you use fast convergence version.

2. Run main_S3TTV.m

Our Reference

If you use this code, please cite the following paper:

@ARTICLE{takemoto2025spatiospectral,
  author={Takemoto, Shingo and Naganuma, Kazuki and Ono, Shunsuke},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Spatio-Spectral Structure Tensor Total Variation for Hyperspectral Image Denoising and Destriping}, 
  year={2025},
  volume={18},
  number={},
  pages={19157-19175},
  doi={10.1109/JSTARS.2025.3586779}
}

Comparison with existing methods

This repository also supports comparison with several existing denoising and destriping methods.

QRNN3D

To evaluate QRNN3D [1], follow the steps below:

1. Download the official code from https://github.com/Vandermode/QRNN3D

2. Download the fine-tuned checkpoint (Pavia Centre) from Google Drive

3. Run hsi_test.py.

Other conventional methods

To compare with other methods (SSTV [2], HSSTV [3], l0-l1HTV [4], STV [5], SSST [6], LRTDTV [7], FGSLR [8], TPTV [9], and FastHyMix [10]):

1. Download and extract the following repositories, and place each extracted folder into the compared_methods/ directory:

   • LRTDTV
     https://github.com/zhaoxile/Hyperspectral-Image-Restoration-via-Total-Variation-Regularized-Low-rank-Tensor-Decomposition

   • FGSLR
     https://chenyong1993.github.io/yongchen.github.io/

   • TPTV
     https://github.com/chuchulyf/ETPTV

   • FastHyMix
     https://github.com/LinaZhuang/HSI-MixedNoiseRemoval-FastHyMix

2. Run main_with_comparisons.m

• Choose the target image (JasperRidge, PaviaUniversity, or Beltsville)
• Enable each method by setting "enable" to true in the corresponding section
• Adjust parameters for each method

3. Run the result visualization script: plot_result.m

• Use show_band to select the band index for visualization
• The script compares results in result/<condition>/<name_method>/ and selects the best result for each method

---

References

[1] K. Wei, Y. Fu, and H. Huang, ``3-D quasi-recurrent neural network for hyperspectral image denoising,'' IEEE Trans. Neural Netw. Learn. Syst., vol. 32, no. 1, pp. 363--375, 2021.

[2] H. K. Aggarwal and A. Majumdar, ``Hyperspectral image denoising using spatio-spectral total variation,'' IEEE Geosci. Remote Sens. Lett., vol. 13, no. 3, pp. 442--446, 2016.

[3] S. Takeyama, S. Ono, and I. Kumazawa, ``A constrained convex optimization approach to hyperspectral image restoration with hybrid spatio-spectral regularization,'' Remote Sens., vol. 12, no. 21, 2020.

[4] M. Wang, Q. Wang, J. Chanussot, and D. Hong, ``$l_0$-$l_1$ hybrid total variation regularization and its applications on hyperspectral image mixed noise removal and compressed sensing,'' IEEE Trans. Geosci. Remote Sens., vol. 59, no. 9, pp. 7695--7710, 2021.

[5] S. Lefkimmiatis, A. Roussos, P. Maragos, and M. Unser, ``Structure tensor total variation,'' SIAM J. Imag. Sci., vol. 8, no. 2, pp. 1090--1122, 2015.

[6] R. Kurihara, S. Ono, K. Shirai, and M. Okuda, ``Hyperspectral image restoration based on spatio-spectral structure tensor regularization,'' in Proc. Eur. Signal Process. Conf. (EUSIPCO), 2017, pp. 488--492.

[7] Y. Wang, J. Peng, Q. Zhao, Y. Leung, X. Zhao, and D. Meng, ``Hyperspectral image restoration via total variation regularized low-rank tensor decomposition,'' IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 11, no. 4, pp. 1227--1243, 2018.

[8] Y. Chen, T. Huang, W. He, X. Zhao, H. Zhang, and J. Zeng, ``Hyperspectral image denoising using factor group sparsity-regularized nonconvex low-rank approximation,'' IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1--16, 2022.

[9] Y. Chen, W. Cao, L. Pang, J. Peng, and X. Cao, ``Hyperspectral image denoising via texture-preserved total variation regularizer,'' IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1--14, 2023.

[10] L. Zhuang and M. K. Ng, ``FastHyMix: Fast and parameter-free hyperspectral image mixed noise removal,'' IEEE Trans. Neural Netw. Learn. Syst., vol. 34, no. 8, pp. 4702--4716, 2023.