Hyperspectral Image Denoising via Group Sparsity Regularized Hybrid Spatio-Spectral Total Variation
Round 1
Reviewer 1 Report
The paper proposes a new hyperspectral image (HSI) denoising model with the group sparsity regularized hybrid spatial-spectral total variation (GHSSTV) and low-rank tensor decomposition. This paper exploits the global low-rank correlations among all modes are explored by the Tucker decomposition, and we propose GHSSTV is designed to avoid over-smoothing that employs the group sparsity in the first-order gradient domain and the second-order ones along the spatial-spectral dimensions. The line of research is highly relevant and the proposal is very promising, nevertheless, I recommend the paper to be majorly revised considering the following comments that are mainly related to the technical contribution and the strength of the simulations.
(1) The technical contribution of the paper is weak with respect to the previous works published in [1-3]. Compared with the smoothness and low-rankness constraints along the spatial and spectral domains in [1-3], the paper also employs the group sparsity along spatial and spectral domains. Thus, I highly suggest the authors give more analysis and necessary discussion with previous works [1-3], which can further interpret that group sparsity is more reasonable than traditional representation in [1-3], otherwise it is hard to highlight the contribution of this paper.
[1] Wang, Y.; Peng, J.; Zhao, Q.; Leung, Y.; Zhao, X. L.; Meng, D.; Hyperspectral image restoration via total variation regularized low-rank tensor decomposition. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 2017, 11, 1227–1243.
[2] Chen, Y.; He, W.; Yokoya, N.; Huang, T. Z. Hyperspectral image restoration using weighted group sparsity-regularized low-rank tensor decomposition. IEEE Trans. Cybern. 2019, 50, 3556–3570.
[3] Takeyama, S.; Ono, S.; Kumazawa, I.; A constrained convex optimization approach to hyperspectral image restoration with hybrid spatio-spectral regularization. RS. 2020, 12, 3541.
(2) For HSI denoising problem, the spatial nonlocal similarity prior has demonstrated powerful characterization ability for noise remove. And for the low-rank property, besides the tensor-based Tucker decomposition, the CP model and the improved tensor decompositions [4-7] have been proposed in recent years. Thus, I suggest the authors to add some discussion about nonlocal tensor decomposition regularized HSI denoising/super-resolution [4-5] and recent some low-rank tensor models [6-7], which can provide some theory supports in terms of contributions.
[4] Zhang, Hongyan, et al. "Hyperspectral image denoising with total variation regularization and nonlocal low-rank tensor decomposition." IEEE Transactions on Geoscience and Remote Sensing 58.5 (2019): 3071-3084.
[5] Xue, Jize, et al. "Spatial-spectral structured sparse low-rank representation for hyperspectral image super-resolution." IEEE Transactions on Image Processing 30 (2021): 3084-3097
[6] Li, Shutao, et al. "Fusing hyperspectral and multispectral images via coupled sparse tensor factorization." IEEE Transactions on Image Processing 27.8 (2018): 4118-4130.
[7] Zeng H, Xie X, Cui H, et al. Hyperspectral Image Restoration via Global L1-2 Spatial–Spectral Total Variation Regularized Local Low-Rank Tensor Recovery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 59(4): 3309-3325.
(3) The notations are not clearly described. There are various grammatical errors and uncommon terminology. To make the notations of the variables clear, the authors should add a Table to list usually used notations.
(4) The authors should include the computational complexity and executing time comparison.
(5) In the experimental section, the authors only test effectiveness of the proposed algorithm by the simulated and real noise. For the denoised HSI, one applied the HSI to subsequent applications, such as object detection and classification, thus I suggest the authors to evaluate the identify characteristics reservation ability of proposed method by testing the classification on AVIRIS Indian Pines Data Set.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
It's an interesting paper on an interesting topic. The writing is a bit too dense and, about the introduction (that gives a state of the art), it should be better having a global evaluation of the class of approaches (showing why such a class doesn't provide results that reaches given goals) instead of having a list of papers with one line comment.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper proposes a group sparsity regularized hybrid spatio-spectral total variation (GHSSTV) and low-rank tensor decomposition for HSI denoising. The group sparsity regularization is used to constrain the hybrid spatio-spectral gradient map, and the low-rank prior is described by Tucker decomposition. Experimental results show the effectiveness of the proposed method. However, this reviewer thinks the novelty of this submission is below the bar for RS. The central part of the algorithm, i.e., the group sparsity to the gradient map, low-rank Tucker decomposition, and HSSTV, is proposed in other works as described in the paper [26][34][35]. Of course, there is nothing wrong with using the combination of these techniques and models. However, the problem is that there is nothing new or unique innovation. Therefore, I must reject it.
Besides, I have a few more concerns about the paper:
The authors may remove the experimental comparison with BM3D and TDL, which are designed for Gaussian noise, not especially for mixed noise.
The parameter selection should be given.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have addressed my comments，the manuscript can be considered to publish.
Reviewer 3 Report
NA
