Self-supervised Hyperspectral Image Restoration using Separable Image Prior

Masahiro Okuda; Ryuji Imamura; Tatsuki Itasaka

arxiv: 1907.00651 · v1 · pith:YDXXUTKWnew · submitted 2019-07-01 · 📡 eess.IV · cs.CV

Self-supervised Hyperspectral Image Restoration using Separable Image Prior

Ryuji Imamura , Tatsuki Itasaka , Masahiro Okuda This is my paper

Pith reviewed 2026-05-25 11:34 UTC · model grok-4.3

classification 📡 eess.IV cs.CV

keywords self-supervisedhyperspectralimage restorationseparable convolutiondenoisingimage priorsingle image

0 comments

The pith

A self-supervised method restores hyperspectral images from a single degraded example by training on synthetic pairs with a separable network.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that hyperspectral image restoration can be done in a self-supervised way by automatically generating training data from just one degraded image. It trains a denoising network using a separable convolutional structure to learn the image prior without needing any clean images. This approach addresses the difficulty of collecting large hyperspectral datasets and the high computational cost of processing many spectral bands. A sympathetic reader would care because it opens up restoration for applications like remote sensing where clean data is scarce. If the claim holds, it means effective denoising is possible in data-limited scenarios.

Core claim

The central claim is that by creating training pairs through synthetic degradations on a single degraded hyperspectral image and training a separable convolutional denoising network, the method acquires the prior of the hyperspectral image and realizes efficient restoration, with experiments demonstrating better performance than state-of-the-art methods.

What carries the argument

Separable convolutional layer that separates the processing to acquire the hyperspectral image prior efficiently.

Load-bearing premise

The synthetic degradations on a single degraded hyperspectral image create training pairs that have the right statistics to learn the clean image prior.

What would settle it

If the restored images from the self-supervised network show no improvement or worse quality than those from a network trained on actual clean hyperspectral data when evaluated on held-out test images with known ground truth.

Figures

Figures reproduced from arXiv: 1907.00651 by Masahiro Okuda, Ryuji Imamura, Tatsuki Itasaka.

**Figure 3.** Figure 3: Four plots indicate histogram of difference of adjac [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** Figure 4: Architecture: We use a simple separable CNN composed [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 5.** Figure 5: Self-supervised hole-filling: (a) original image, ( [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗

**Figure 6.** Figure 6: Examples of corrupted images in mixed noise removal [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗

**Figure 7.** Figure 7: Real Noise Removal tensor decomposition. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 11(4):1227–1243, 2018. [31] Q. Yuan, L. Zhang, and H. Shen. Hyperspectral image denoising employing a spectral-spatial adaptive total variation model. IEEE Trans. Geosci. Remote Sens., 50(10):3660–3677, 2012. [32] H. Zhang. Hyperspectral image denoising with cubic total variation model… view at source ↗

read the original abstract

Supervised learning with a convolutional neural network is recognized as a powerful means of image restoration. However, most such methods have been designed for application to grayscale and/or color images; therefore, they have limited success when applied to hyperspectral image restoration. This is partially owing to large datasets being difficult to collect, and also the heavy computational load associated with the restoration of an image with many spectral bands. To address this difficulty, we propose a novel self-supervised learning strategy for application to hyperspectral image restoration. Our method automatically creates a training dataset from a single degraded image and trains a denoising network without any clear images. Another notable feature of our method is the use of a separable convolutional layer. We undertake experiments to prove that the use of a separable network allows us to acquire the prior of a hyperspectral image and to realize efficient restoration. We demonstrate the validity of our method through extensive experiments and show that our method has better characteristics than those that are currently regarded as state-of-the-art.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives a self-supervised HSI denoising method that builds training pairs from one already-degraded image plus separable convolutions, but offers no evidence the synthetic pairs recover an unbiased clean prior.

read the letter

The core idea is straightforward: take one noisy hyperspectral image, apply extra synthetic degradations to make noisy-clean pairs on the fly, then train a separable CNN to learn the image prior without ever seeing clean data. The separable layers are meant to keep computation manageable across dozens of spectral bands. That combination is the actual novelty here, and it directly targets the data scarcity problem that supervised CNNs face in hyperspectral work. If the experiments really show gains over existing methods, that part is useful for anyone who has to restore sensor data without paired references. The efficiency claim also makes sense given the band count. The soft spot sits right at the method's center. The training pairs come from further degrading an input that is already degraded, so the learned prior is conditioned on whatever artifacts and noise are already present. Nothing in the abstract or the stress-test description shows why the composed degradation operator preserves the statistics needed to recover the true clean image. A mismatch in noise model or spectral correlation would bias the result, and the paper supplies no derivation, bound, or ablation that addresses this. The claim of better characteristics than SOTA is stated but not backed by numbers or error analysis in the available text, which leaves the practical payoff unclear. This is for people working on hyperspectral restoration pipelines who need to operate without large clean datasets. It is narrow in scope but addresses a real bottleneck. The work shows clear thinking on the data problem and the architecture choice, so it deserves a serious referee even if the validation on the pair-generation step needs tightening. I would send it to review and ask specifically for the degradation model details and quantitative metrics.

Referee Report

2 major / 0 minor

Summary. The paper proposes a self-supervised strategy for hyperspectral image (HSI) restoration that automatically generates training pairs from a single degraded input image via additional synthetic degradations, trains a separable CNN to learn the HSI prior, and claims superior restoration performance over current state-of-the-art methods without requiring any clean reference images.

Significance. If the central claim holds, the work would be significant for HSI restoration tasks where clean training data are scarce, as it enables learning from a single degraded observation and uses separable convolutions for computational efficiency on high-dimensional spectral data.

major comments (2)

[Abstract] Abstract: the claim that 'extensive experiments' demonstrate better characteristics than SOTA provides no quantitative metrics, error bars, or details on how the self-supervised pairs are generated, leaving the central empirical support unverifiable from the text.
[Abstract] Abstract (self-supervised strategy paragraph): the assumption that synthetic degradations applied to an already-degraded HSI produce training pairs whose joint distribution matches real clean-to-degraded observations is stated without derivation, bound, or validation; any mismatch in noise model, spectral correlation, or pre-existing artifacts would bias the learned prior away from the clean image statistics.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below, indicating the revisions we will make to strengthen the presentation of our claims.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that 'extensive experiments' demonstrate better characteristics than SOTA provides no quantitative metrics, error bars, or details on how the self-supervised pairs are generated, leaving the central empirical support unverifiable from the text.

Authors: We agree that the abstract, being a high-level summary, does not include specific quantitative metrics, error bars, or details on pair generation, which limits immediate verifiability of the empirical claims. The full manuscript provides these details in the experiments section. To address the concern, we will revise the abstract to incorporate key quantitative results (e.g., representative PSNR/SSIM values with comparisons) and a brief description of the self-supervised pair synthesis process. This will be a targeted update to the abstract only. revision: yes
Referee: [Abstract] Abstract (self-supervised strategy paragraph): the assumption that synthetic degradations applied to an already-degraded HSI produce training pairs whose joint distribution matches real clean-to-degraded observations is stated without derivation, bound, or validation; any mismatch in noise model, spectral correlation, or pre-existing artifacts would bias the learned prior away from the clean image statistics.

Authors: The referee correctly notes that the abstract states the self-supervised pair generation approach without providing a derivation, bound, or explicit validation of the joint distribution assumption. The manuscript focuses on the empirical outcomes rather than a formal proof of distribution equivalence. We will expand the method description in the revised manuscript to include a discussion of the underlying assumptions, potential mismatches (e.g., in noise models or spectral correlations), and supporting empirical checks, while acknowledging this as a limitation of the current theoretical analysis. revision: yes

Circularity Check

0 steps flagged

No significant circularity; self-supervised pair generation is empirical, not definitional

full rationale

The paper's core claim is an empirical self-supervised procedure: synthetic degradations are applied to one input degraded HSI to form training pairs, a separable CNN is trained on those pairs, and the resulting network is applied for restoration. No equation or step is shown to reduce the output prior or restored image to the input degradation by algebraic identity or by renaming a fitted parameter. The abstract and description contain no load-bearing self-citations, imported uniqueness theorems, or ansatzes smuggled via prior work. The method is presented as a practical training recipe whose validity is asserted via experiments rather than forced by construction from its own inputs. This is the common honest case of a non-circular empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the premise that a network trained via self-supervision on synthetically degraded versions of one input can learn a general hyperspectral image prior; no free parameters, axioms, or invented entities are explicitly introduced in the abstract.

pith-pipeline@v0.9.0 · 5705 in / 1098 out tokens · 19090 ms · 2026-05-25T11:34:41.591553+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · 3 internal anchors

[1]

H. K. Aggarwal and A. Majumdar. Hyperspectral image deno ising using spatio-spectral total variation. IEEE Geosci. Remote Sens. Lett. , 13(3):442–446, 2016

work page 2016
[2]

J. M. Bioucas-Dias and J. M. Nascimento. Hyperspectral s ubspace identiﬁcation. IEEE Transactions on Geoscience and Remote Sensing , 46(8):2435–2445, 2008

work page 2008
[3]

J. M. Bioucas-Dias, A. Plaza, G. Camps-V alls, P . Scheund ers, N. Nasrabadi, and J. Chanussot. Hyperspectral remote sensi ng data analysis and future challenges. IEEE Geoscience and remote sensing magazine, 1(2):6–36, 2013

work page 2013
[4]

J. M. Bioucas-Dias, A. Plaza, N. Dobigeon, M. Parente, Q. Du, P . Gader, and J. Chanussot. Hyperspectral unmixing overview: Geomet rical, statis- tical, and sparse regression-based approaches. IEEE journal of selected topics in applied earth observations and remote sensing , 5(2):354–379, 2012

work page 2012
[5]

H. C. Burger, C. J. Schuler, and S. Harmeling. Image denoi sing: Can plain neural networks compete with bm3d? In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on , pages 2392–

work page 2012
[6]

Chen and T

Y . Chen and T. Pock. Trainable nonlinear reaction diffus ion: A ﬂexible framework for fast and effective image restoration. IEEE transactions on pattern analysis and machine intelligence , 39(6):1256–1272, 2017

work page 2017
[7]

F. Chollet. Xception: Deep learning with depthwise sepa rable convolu- tions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1251–1258, 2017

work page 2017
[8]

W. He, H. Zhang, L. Zhang, and H. Shen. Total-variation-r egularized low-rank matrix factorization for hyperspectral image res toration. IEEE Trans. Geosci. Remote Sens. , 54(1):178–188, 2016

work page 2016
[9]

D. P . Kingma and J. Ba. Adam: A method for stochastic optim ization. arXiv preprint arXiv:1412.6980 , 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014
[10]

Noise2Void - Learning Denoising from Single Noisy Images

A. Krull, T.-O. Buchholz, and F. Jug. Noise2void-learn ing denoising from single noisy images. arXiv preprint arXiv:1811.10980 , 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[11]

Kurihara, S

R. Kurihara, S. Ono, K. Shirai, and M. Okuda. Hyperspect ral image restoration based on spatio-spectral structure tensor reg ularization. In Proc. Eur . Signal Process. Conf. , pages 488–492, 2017

work page 2017
[12]

Lefkimmiatis, A

S. Lefkimmiatis, A. Roussos, P .Maragos, and M. Unser. S tructure tensor total variation. SIAM J. Imag. Sci. , 8(2):1090–1122, 2015

work page 2015
[13]

Noise2Noise: Learning Image Restoration without Clean Data

J. Lehtinen, J. Munkberg, J. Hasselgren, S. Laine, T. Ka rras, M. Aittala, and T. Aila. Noise2noise: Learning image restoration witho ut clean data. arXiv preprint arXiv:1803.04189 , 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[14]

Maggioni, V

M. Maggioni, V . Katkovnik, K. Egiazarian, and A. Foi. No nlocal transform-domain ﬁlter for volumetric data denoising and r econstruction. IEEE Trans. Image Process. , 22(1):119–133, 2013

work page 2013
[15]

S. Nie, L. Gu, Y . Zheng, A. Lam, N. Ono, and I. Sato. Deeply learned ﬁlter response functions for hyperspectral reconstructio n. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogn ition, pages 4767–4776, 2018

work page 2018
[16]

Okabe and M

T. Okabe and M. Okuda. Computationally efﬁcient reﬂect ance es- timation for hyperspectral images. IEICE Trans. on info. Sys. , E100.D(9):2253–2256, 2017

work page 2017
[17]

S. Ono, K. Shirai, and M. Okuda. V ectorial total variati on based on arranged structure tensor for multichannel image restorat ion. In Proc. IEEE Int. Conf. Acoust. Speech Signal Process. , pages 4528–4532, 2016

work page 2016
[18]

Plaza, J

A. Plaza, J. A. Benediktsson, J. W. Boardman, J. Brazile , L. Bruzzone, G. Camps-V alls, J. Chanussot, M. Fauvel, P . Gamba, A. Gualti eri, et al. Recent advances in techniques for hyperspectral imag e processing. Remote sensing of environment , 113:S110–S122, 2009

work page 2009
[19]

V . B. S. Prasath, D. V orotnikov, R. Pelapur, S. Jose, G. S eetharaman, and K. Palaniappan. Multiscale tikhonov-total variation imag e restoration using spatially varying edge coherence exponent. IEEE Trans. Image Process., 24(12):5220–5235, 2015

work page 2015
[20]

Y . Qu, H. Qi, and C. Kwan. Unsupervised sparse dirichlet -net for hyperspectral image super-resolution. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 2511– 2520, 2018

work page 2018
[21]

Rizkinia, T

M. Rizkinia, T. Baba, K. Shirai, and M. Okuda. Local spec tral component decomposition for multi-channel image denoisin g. IEEE Trans. Image Process. , 25, Issue:7:3208–3218, July 2016

work page 2016
[22]

Rizkinia and M

M. Rizkinia and M. Okuda. Joint local abundance sparse u n- mixing for hyperspectral images. Remote Sensing , 9(12):1224; doi:10.3390/rs9121224, 2017

work page doi:10.3390/rs9121224 2017
[23]

Schmidt and S

U. Schmidt and S. Roth. Shrinkage ﬁelds for effective im age restoration. In Proceedings of the IEEE Conference on Computer Vision and Pa ttern Recognition, pages 2774–2781, 2014

work page 2014
[24]

Shaw and D

G. Shaw and D. Manolakis. Signal processing for hypersp ectral image exploitation. IEEE Signal Process. Magazine , 19(1):12–16, 2002

work page 2002
[25]

Z. Shi, C. Chen, Z. Xiong, D. Liu, and F. Wu. Hscnn+: Advan ced cnn- based hyperspectral recovery from rgb images. In IEEE/CVF Conference on Computer Vision and Pattern Recognition W orkshops (CVPR W), volume 3, page 5, 2018

work page 2018
[26]

Shocher, N

A. Shocher, N. Cohen, and M. Irani. zero-shot super-res olution using deep internal learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 3118–3126, 2018

work page 2018
[27]

D. W. Stein, S. G. Beaven, L. E. Hoff, E. M. Winter, A. P . Sc haum, and A. D. Stocker. Anomaly detection from hyperspectral imager y. IEEE signal processing magazine , 19(1):58–69, 2002

work page 2002
[28]

Takeyama, S

S. Takeyama, S. Ono, and I. Kumazawa. Hyperspectral ima ge restoration by hybrid spatio-spectral total variation. In Proc. IEEE Int. Conf. Acoust. Speech Signal Process. , pages 4586–4590, 2017

work page 2017
[29]

Ulyanov, A

D. Ulyanov, A. V edaldi, and V . Lempitsky. Deep image pri or. In Proceedings of the IEEE Conference on Computer Vision and Pa ttern Recognition, pages 9446–9454, 2018

work page 2018
[30]

Y . Wang, J. Peng, Q. Zhao, Y . Leung, X.-L. Zhao, and D. Men g. Hyperspectral image restoration via total variation regul arized low-rank JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2019 8 Original DIP N2V Ours Fig. 7. Real Noise Removal tensor decomposition. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 11(4...

work page 2019
[31]

Y uan, L

Q. Y uan, L. Zhang, and H. Shen. Hyperspectral image deno ising employing a spectral-spatial adaptive total variation mod el. IEEE Trans. Geosci. Remote Sens. , 50(10):3660–3677, 2012

work page 2012
[32]

H. Zhang. Hyperspectral image denoising with cubic tot al variation model. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci. , I-7:95– 98, 2012

work page 2012
[33]

Zhang, W

H. Zhang, W. He, L. Zhang, H. Shen, and Q. Y uan. Hyperspec tral image restoration using low-rank matrix recovery. IEEE Trans. Geosci. Remote Sens., 52(8):4729–4743, 2014

work page 2014
[34]

Zhang, J

H. Zhang, J. Li, Y . Huang, and L. Zhang. A nonlocal weight ed joint sparse representation classiﬁcation method for hyperspec tral imagery. IEEE J. Sel. Topics Appl. Earth Obs. Remote Sens. , 7(6):2056–2065, 2014

work page 2056
[35]

Zhang, W

K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang. Beyond a g aussian denoiser: Residual learning of deep cnn for image denoising . IEEE Transactions on Image Processing , 26(7):3142–3155, 2017

work page 2017
[36]

Zhuang and J

L. Zhuang and J. M. Bioucas-Dias. Fast hyperspectral im age denoising and inpainting based on low-rank and sparse representation s. IEEE Journal of Selected Topics in Applied Earth Observations an d Remote Sensing, 11(3):730–742, 2018

work page 2018

[1] [1]

H. K. Aggarwal and A. Majumdar. Hyperspectral image deno ising using spatio-spectral total variation. IEEE Geosci. Remote Sens. Lett. , 13(3):442–446, 2016

work page 2016

[2] [2]

J. M. Bioucas-Dias and J. M. Nascimento. Hyperspectral s ubspace identiﬁcation. IEEE Transactions on Geoscience and Remote Sensing , 46(8):2435–2445, 2008

work page 2008

[3] [3]

J. M. Bioucas-Dias, A. Plaza, G. Camps-V alls, P . Scheund ers, N. Nasrabadi, and J. Chanussot. Hyperspectral remote sensi ng data analysis and future challenges. IEEE Geoscience and remote sensing magazine, 1(2):6–36, 2013

work page 2013

[4] [4]

J. M. Bioucas-Dias, A. Plaza, N. Dobigeon, M. Parente, Q. Du, P . Gader, and J. Chanussot. Hyperspectral unmixing overview: Geomet rical, statis- tical, and sparse regression-based approaches. IEEE journal of selected topics in applied earth observations and remote sensing , 5(2):354–379, 2012

work page 2012

[5] [5]

H. C. Burger, C. J. Schuler, and S. Harmeling. Image denoi sing: Can plain neural networks compete with bm3d? In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on , pages 2392–

work page 2012

[6] [6]

Chen and T

Y . Chen and T. Pock. Trainable nonlinear reaction diffus ion: A ﬂexible framework for fast and effective image restoration. IEEE transactions on pattern analysis and machine intelligence , 39(6):1256–1272, 2017

work page 2017

[7] [7]

F. Chollet. Xception: Deep learning with depthwise sepa rable convolu- tions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1251–1258, 2017

work page 2017

[8] [8]

W. He, H. Zhang, L. Zhang, and H. Shen. Total-variation-r egularized low-rank matrix factorization for hyperspectral image res toration. IEEE Trans. Geosci. Remote Sens. , 54(1):178–188, 2016

work page 2016

[9] [9]

D. P . Kingma and J. Ba. Adam: A method for stochastic optim ization. arXiv preprint arXiv:1412.6980 , 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014

[10] [10]

Noise2Void - Learning Denoising from Single Noisy Images

A. Krull, T.-O. Buchholz, and F. Jug. Noise2void-learn ing denoising from single noisy images. arXiv preprint arXiv:1811.10980 , 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[11] [11]

Kurihara, S

R. Kurihara, S. Ono, K. Shirai, and M. Okuda. Hyperspect ral image restoration based on spatio-spectral structure tensor reg ularization. In Proc. Eur . Signal Process. Conf. , pages 488–492, 2017

work page 2017

[12] [12]

Lefkimmiatis, A

S. Lefkimmiatis, A. Roussos, P .Maragos, and M. Unser. S tructure tensor total variation. SIAM J. Imag. Sci. , 8(2):1090–1122, 2015

work page 2015

[13] [13]

Noise2Noise: Learning Image Restoration without Clean Data

J. Lehtinen, J. Munkberg, J. Hasselgren, S. Laine, T. Ka rras, M. Aittala, and T. Aila. Noise2noise: Learning image restoration witho ut clean data. arXiv preprint arXiv:1803.04189 , 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[14] [14]

Maggioni, V

M. Maggioni, V . Katkovnik, K. Egiazarian, and A. Foi. No nlocal transform-domain ﬁlter for volumetric data denoising and r econstruction. IEEE Trans. Image Process. , 22(1):119–133, 2013

work page 2013

[15] [15]

S. Nie, L. Gu, Y . Zheng, A. Lam, N. Ono, and I. Sato. Deeply learned ﬁlter response functions for hyperspectral reconstructio n. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogn ition, pages 4767–4776, 2018

work page 2018

[16] [16]

Okabe and M

T. Okabe and M. Okuda. Computationally efﬁcient reﬂect ance es- timation for hyperspectral images. IEICE Trans. on info. Sys. , E100.D(9):2253–2256, 2017

work page 2017

[17] [17]

S. Ono, K. Shirai, and M. Okuda. V ectorial total variati on based on arranged structure tensor for multichannel image restorat ion. In Proc. IEEE Int. Conf. Acoust. Speech Signal Process. , pages 4528–4532, 2016

work page 2016

[18] [18]

Plaza, J

A. Plaza, J. A. Benediktsson, J. W. Boardman, J. Brazile , L. Bruzzone, G. Camps-V alls, J. Chanussot, M. Fauvel, P . Gamba, A. Gualti eri, et al. Recent advances in techniques for hyperspectral imag e processing. Remote sensing of environment , 113:S110–S122, 2009

work page 2009

[19] [19]

V . B. S. Prasath, D. V orotnikov, R. Pelapur, S. Jose, G. S eetharaman, and K. Palaniappan. Multiscale tikhonov-total variation imag e restoration using spatially varying edge coherence exponent. IEEE Trans. Image Process., 24(12):5220–5235, 2015

work page 2015

[20] [20]

Y . Qu, H. Qi, and C. Kwan. Unsupervised sparse dirichlet -net for hyperspectral image super-resolution. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 2511– 2520, 2018

work page 2018

[21] [21]

Rizkinia, T

M. Rizkinia, T. Baba, K. Shirai, and M. Okuda. Local spec tral component decomposition for multi-channel image denoisin g. IEEE Trans. Image Process. , 25, Issue:7:3208–3218, July 2016

work page 2016

[22] [22]

Rizkinia and M

M. Rizkinia and M. Okuda. Joint local abundance sparse u n- mixing for hyperspectral images. Remote Sensing , 9(12):1224; doi:10.3390/rs9121224, 2017

work page doi:10.3390/rs9121224 2017

[23] [23]

Schmidt and S

U. Schmidt and S. Roth. Shrinkage ﬁelds for effective im age restoration. In Proceedings of the IEEE Conference on Computer Vision and Pa ttern Recognition, pages 2774–2781, 2014

work page 2014

[24] [24]

Shaw and D

G. Shaw and D. Manolakis. Signal processing for hypersp ectral image exploitation. IEEE Signal Process. Magazine , 19(1):12–16, 2002

work page 2002

[25] [25]

Z. Shi, C. Chen, Z. Xiong, D. Liu, and F. Wu. Hscnn+: Advan ced cnn- based hyperspectral recovery from rgb images. In IEEE/CVF Conference on Computer Vision and Pattern Recognition W orkshops (CVPR W), volume 3, page 5, 2018

work page 2018

[26] [26]

Shocher, N

A. Shocher, N. Cohen, and M. Irani. zero-shot super-res olution using deep internal learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 3118–3126, 2018

work page 2018

[27] [27]

D. W. Stein, S. G. Beaven, L. E. Hoff, E. M. Winter, A. P . Sc haum, and A. D. Stocker. Anomaly detection from hyperspectral imager y. IEEE signal processing magazine , 19(1):58–69, 2002

work page 2002

[28] [28]

Takeyama, S

S. Takeyama, S. Ono, and I. Kumazawa. Hyperspectral ima ge restoration by hybrid spatio-spectral total variation. In Proc. IEEE Int. Conf. Acoust. Speech Signal Process. , pages 4586–4590, 2017

work page 2017

[29] [29]

Ulyanov, A

D. Ulyanov, A. V edaldi, and V . Lempitsky. Deep image pri or. In Proceedings of the IEEE Conference on Computer Vision and Pa ttern Recognition, pages 9446–9454, 2018

work page 2018

[30] [30]

Y . Wang, J. Peng, Q. Zhao, Y . Leung, X.-L. Zhao, and D. Men g. Hyperspectral image restoration via total variation regul arized low-rank JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2019 8 Original DIP N2V Ours Fig. 7. Real Noise Removal tensor decomposition. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 11(4...

work page 2019

[31] [31]

Y uan, L

Q. Y uan, L. Zhang, and H. Shen. Hyperspectral image deno ising employing a spectral-spatial adaptive total variation mod el. IEEE Trans. Geosci. Remote Sens. , 50(10):3660–3677, 2012

work page 2012

[32] [32]

H. Zhang. Hyperspectral image denoising with cubic tot al variation model. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci. , I-7:95– 98, 2012

work page 2012

[33] [33]

Zhang, W

H. Zhang, W. He, L. Zhang, H. Shen, and Q. Y uan. Hyperspec tral image restoration using low-rank matrix recovery. IEEE Trans. Geosci. Remote Sens., 52(8):4729–4743, 2014

work page 2014

[34] [34]

Zhang, J

H. Zhang, J. Li, Y . Huang, and L. Zhang. A nonlocal weight ed joint sparse representation classiﬁcation method for hyperspec tral imagery. IEEE J. Sel. Topics Appl. Earth Obs. Remote Sens. , 7(6):2056–2065, 2014

work page 2056

[35] [35]

Zhang, W

K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang. Beyond a g aussian denoiser: Residual learning of deep cnn for image denoising . IEEE Transactions on Image Processing , 26(7):3142–3155, 2017

work page 2017

[36] [36]

Zhuang and J

L. Zhuang and J. M. Bioucas-Dias. Fast hyperspectral im age denoising and inpainting based on low-rank and sparse representation s. IEEE Journal of Selected Topics in Applied Earth Observations an d Remote Sensing, 11(3):730–742, 2018

work page 2018