Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Model-Guided Network with Cluster-Based Operators for Spatio-Spectral Super-Resolution

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a fully unfolded, end-to-end network that decomposes spatio-spectral super-resolution into spatial super-resolution, spectral super-resolution, and fusion outperforms existing joint and composed methods across three…

desk verdict A solid empirical SSSR paper with a genuinely new decomposition–unfolding combination, but the cluster-based operator claim is undercut by an unspecified gradient path that needs fixing. read the letter →

arxiv 2505.24605 v1 pith:F2NAXYC4 submitted 2025-05-30 eess.IV cs.CV

Ivan Pereira-Sánchez , Julia Navarro , Ana Belén Petro , Joan Duran This is my paper
classification eess.IVcs.CV
keywords jointspatio-spectralsuper-resolutionspectralspatialimagefusionalgorithmunfoldinghyperspectralimagingcluster-basedoperatorsback-projection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes an end-to-end, model-driven network that reconstructs a high-resolution hyperspectral image from a single low-resolution multispectral image by explicitly decomposing the problem into spatial super-resolution, spectral super-resolution, and fusion. Each subtask is built by unfolding a proximal-gradient variational scheme and replacing its operators with learnable modules, including a back-projection-inspired spatial upsampler and cluster-based spectral operators. The authors report that the method achieves the best or second-best PSNR, SSIM, SAM, and ERGAS scores on CAVE, Pavia Center, and NTIRE2020 at sampling factors 2, 4, and 8, compared with existing joint super-resolution methods and composed single-task baselines. If correct, this would give a single-model alternative to paired-capture fusion systems for hyperspectral imaging.

What carries the argument

The load-bearing mechanism is the unfolded proximal-gradient step $u^{n+1} = \mathrm{prox}_{\tau\lambda R}(u^n - \tau \nabla D(u^n))$, where each subproblem's fidelity gradient and proximity operator are replaced by learnable modules. For spectral reconstruction, the key object is a pair of cluster-based operators: a learned clustering module (Softmax followed by argmax over $M=10$ clusters) assigns each pixel to a cluster, and each cluster gets its own MLP mapping between MS and HS space, so spectral transforms vary by spectral response class. For spatial upsampling, the key object is a back-projection step $u^{k+1} = u^k + \kappa * \mathrm{Bic}(DB(u^k) - f)$ emulated with learnable Down and Up operators. For fusion, the key identity is the radiometric constraint $u - u_{\mathrm{SSR}} = (u_{\mathrm{SSR}} \oslash u_{\mathrm{SR}}) \odot (\hat{u}_{\mathrm{SR}} - u_{\mathrm{SR}})$, converted into a fidelity term whose gradient injects HR-MSI geometry. A post-processing module applies multi-head attention over pixel patches restricted to the top 10% most similar neighbors.

What would settle it

Inspect the backward pass through the argmax in the clustering module: if cluster assignments are identical for a randomly initialized and a fully trained network, or if replacing the hard argmax with a soft weighted average changes PSNR by more than the reported gains, then the learned clustering is not the load-bearing mechanism. A simpler test is to train the full model with the clustering module removed and a single global MLP for spectral transforms; if the gap to the reported results on CAVE or Pavia closes, the cluster-based operators are not essential.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that an interpretable unfolded architecture that separates the joint spatio-spectral super-resolution problem into three coupled subproblems—spatial super-resolution, spectral super-resolution, and fusion—can outperform both dedicated SSSR networks and sequential compositions of single-task networks. The novelty rests on three components: a spatial upsampler that emulates one iteration of classical back-projection adapted to arbitrary scaling factors through prime-factor decomposition; spectral upsampling and downsampling operators that partition pixels into learned clusters and apply cluster-specific MLPs; and a fusion fidelity term derived from a radiometric constraint that injects HR-MSI geometry while preserving LR-HSI chromaticity. The paper verifies this claim with quantitative comparisons across three datasets and three sampling ratios, and with ablations isolating each operator.

Load-bearing premise

The cluster-based spectral operators assume that sorting every pixel into one of ten groups, with each group getting its own spectral transform, is something the network can actually learn and is enough to represent spectral variation.

Editorial extensions

If this is right

  • Single-image spatio-spectral super-resolution becomes a practical substitute for dual-sensor fusion when only an LR-MSI is available, since the network learns the spectral mapping and spatial details together.
  • Unfolding gives each stage an interpretable role, and the ablation shows performance peaks at 4 stages and 128 features, indicating that the architecture can be tuned by the number of iterations and feature width.
  • The learned cluster operators imply that ten spectral-response clusters are enough to approximate the MS-to-HS mapping across diverse scenes, as the method generalizes across CAVE, Pavia Center, and NTIRE2020.
  • The post-processing attention restricted to the top 10% neighbors claims to capture long-range dependencies at reduced cost, which could make such nonlocal modules feasible in the high-resolution domain.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the clustering module does not receive gradients through the argmax during training, the cluster assignments may be effectively fixed rather than learned, so the ablation against k-means would compare random fixed groups to k-means rather than a genuinely learned clustering.
  • Running the spatial and spectral branches in parallel suggests a testable extension: enforcing a consistency loss between the intermediate HR-MSI and LR-HSI could further improve fusion, which the paper does not explore.
  • The radiometric fusion constraint is adapted from pansharpening, so the same unfolded constraint could be applied to hypersharpening or other multi-sensor fusion settings where spectral responses are known instead of learned.
  • The gains are largest at the hardest sampling factor of 8 on CAVE and NTIRE2020, hinting that the cluster-based operators and back-projection upsampler are most beneficial in the low-information regime; evaluating at factor 16 would test this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes JSSUNet, an end-to-end unfolding-based framework for joint spatio-spectral super-resolution (SSSR) from a single low-resolution multispectral image. The method decomposes SSSR into spatial SR, spectral SR, and fusion subproblems, each solved by unfolding proximal gradient iterations with learnable operators: a back-projection-inspired upsampler for spatial SR, cluster-based SpecDown/SpecUp operators for spectral SR, and a radiometric-constraint-based fusion with low-frequency estimation and high-frequency injection. An efficient nonlocal multi-head attention post-processing module is added. Experiments on CAVE, Pavia, and NTIRE2020 at sampling factors 2, 4, and 8 compare against SSFIN, US3RN, and three composed baselines, reporting best or second-best PSNR, SSIM, SAM, and ERGAS. Ablations study spatial upsampling variants, spectral cluster assignment, number of stages, number of features, loss hyperparameters, and post-processing.

Significance. If the empirical results hold, the paper offers a useful benchmark and a practical architecture, with a clear decomposition of SSSR into interpretable subproblems. The variational derivations in Eqs. (3)-(5) are algebraically correct; the back-projection upsampling and the efficient top-k attention post-processing are sensible engineering contributions. The evaluation is external (held-out test sets) and includes a reasonably broad set of baselines and datasets, with extensive ablations. The main reservations concern the training of the cluster assignments and the absence of statistical uncertainty quantification.

major comments (3)
  1. [Section III-B and Figure 5] The cluster assignment uses a hard argmax after a Softmax, and the clusters are computed during the initialization step and used to construct SpecDown and SpecUp. No straight-through estimator, soft assignment, or auxiliary loss for the cluster head is described, and the training loss (6) only supervises stage outputs. Because back-propagation through a hard argmax is zero almost everywhere, the cluster-head convolutions do not receive gradients from (6); if the assignments are computed only once at initialization, they are frozen during training. Consequently, Table IV's 'Learned clusters' row does not demonstrate that the clustering itself is learned, even though the cluster-specific MLPs are trained. This weakens a central novelty claim and the interpretability argument; please either use a differentiable assignment (e.g., soft or straight-through), provide evidence that the cluster head is optimized, or revise the claims and the interpretation of Table IV accordingly.
  2. [Section V and Tables II, V, VI, VIII] All experimental results are reported as single-run point estimates. The paper claims 'superiority ... across different datasets and sampling conditions' in the Introduction and Conclusions, but without repeated runs, standard deviations, or significance tests, differences such as CAVE s=4 PSNR 37.40 (Ours) vs 37.02 (SSFIN) or the post-processing gain of +0.15 dB in Table VIII cannot be distinguished from run-to-run variation. Please report the number of seeds and error bars (or a significance analysis) for at least the main comparison table, and soften the superiority claims accordingly.
  3. [Section IV, Equation (6)] The first term is written as l1(uK_Fus, g), comparing the final HR-HSI output to the low-resolution hyperspectral image g. The surrounding text says the loss includes 'an additional term for the final result' and references the HR-HSI G, so this appears to be a typo for l1(uK_Fus, G). If the equation is implemented as printed, the fused output is trained against a target with incompatible spatial dimensions; please correct the equation and confirm the implementation.
minor comments (5)
  1. [Section I and Section V] The claim of a 'comprehensive evaluation comparing all SSSR state-of-the-art methods with publicly available code' is inconsistent with the absence of LISSF [64] in Table II; state explicitly whether LISSF has no public code or include it in the comparison.
  2. [Section III-C] The notation is confusing: uSR and uSSR are first described as the low frequencies of the HR-MSI and the LR-HSI, respectively, but then LFE modules map them to the HS domain for the pixel-wise products in Eq. (5). Please clarify the domain and dimensions of each variable.
  3. [Section VI-E and Table VIII] Calling the post-processing improvement 'significant' is overstated: PSNR improves by only 0.15 dB, and ERGAS slightly worsens (8.27 to 8.28); the text also says 'where the loss is negligible' while reporting a worse ERGAS.
  4. [Section VII] Minor typos: 'has been assesses' should be 'has been assessed'; 'exhibits' should be 'exhibit'; Section VI-D refers to 'Table VI-D' where the table is labeled Table VII; 'hyperparemeters' in Table VII header should be 'hyperparameters'.
  5. [Section V-B] The Pavia test set contains only four crops; reporting metrics aggregated over four crops without variance may be unstable. Consider reporting per-crop results or additional splits.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is trained on data and evaluated on held-out benchmarks, and the unfolded operators are not fitted to or defined from the target outputs.

full rationale

The paper's derivation chain is self-contained and empirically grounded. The SSSR problem is decomposed into SR, SSR, and fusion subtasks, each solved by unfolding the proximal-gradient iteration (2) and replacing the mathematical operators with trainable modules. The fidelity terms are standard observation-model gradients; the learnable modules (Down, BPUp, SpecDown, SpecUp, Prox modules) are optimized on training data and assessed on held-out CAVE, Pavia, and NTIRE2020 test sets. No equation in the paper is fitted to test data, and no predicted quantity is defined in terms of the target variable by construction. The cluster-based spectral operators are an architectural choice: even if the hard-argmax after Softmax blocks gradient flow to the cluster-head and makes the 'learned clusters' effectively random fixed partitions, that is a trainability/novelty concern rather than circularity, because the empirical comparisons in Tables II, III, IV, V, VI, VII, and VIII are external evaluations on unseen images. The self-citations, notably the radiometric constraint from [14] and the survey [42] that includes some of the present authors, are used as modeling inspiration or background context, not as the justification for the reported superiority. The claim of 'first' combination of SR, SSR, and fusion via unfolding is a novelty statement, not a circular derivation. Therefore, there is no identified circular step by the paper's own equations or by load-bearing self-citation.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The method is a supervised deep network, so the 'axioms' are its modeling assumptions rather than mathematical postulates. All operators are trained, so the model assumptions are loose priors, not hard constraints. The free parameters are mostly standard hyperparameters, but several are tuned on one dataset and then applied to others, which is a mild selection risk.

free parameters (5)
  • Number of clusters M = 10
    Section III-B: M=10 chosen in practice; no sensitivity analysis in ablations, though Table IV compares no clusters, k-means, and learned clusters.
  • Number of unfolding stages K = 4
    Section IV and VI-C: K=4 selected as best among 2,4,6,8 on CAVE sampling 4.
  • Number of features in residual blocks = 128
    Section IV and VI-C: 128 features gave best PSNR/SSIM/SAM among 16-256 on CAVE sampling 4.
  • Loss hyperparameters alpha_SR, alpha_SSR, alpha_Fus = {2,0.5,0}, {1,1,0.5}, {0.5,1,1}
    Section VI-D: comb2,1,1 selected as best of 8 combinations on CAVE sampling 4.
  • Patch size P, embedding dim ed, window size, top-k fraction = P=11, ed=8, window 11x11, top 10%
    Section III-D: values set by hand without ablation in this paper.
assumptions (5)
  • domain assumption Spatial observation model f = DB(u) + eta with DB a low-pass filter followed by s-decimation.
    Section III-A, equation (3). Standard for SR, but the actual degradation used to generate training data is not specified.
  • domain assumption Spectral observation model f = S(u) + eta with S a linear spectral downsampling.
    Section III-B. Assumes linear mixing of spectral bands, which the paper itself notes does not strictly hold.
  • standard math Proximity operator approximation prox_{tau lambda R}(x) approx x - tau lambda grad R(x), with residual networks used as prox modules.
    Section III, after equation (2), citing [66]. Requires small tau and differentiable R; in practice the networks are unconstrained.
  • domain assumption Cluster-consistency: SpecDown and SpecUp are constant within clusters of similar spectral response.
    Section III-B: 'we assume that SpecDown and SpecUp should vary for pixels with different spectral responses but remain consistent for those with similar spectral characteristics.'
  • domain assumption Fusion radiometric constraint u - uSSR = (uSSR / uSR) * (uhatSR - uSR), adapted from pansharpening.
    Section III-C, equation (5), from [14]. Assumes a multiplicative relation between low and high frequencies that may not hold for estimated inputs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Model-Guided Network with Cluster-Based Operators for Spatio-Spectral Super-Resolution." pith.science (2026). https://pith.science/paper/F2NAXYC4

@misc{pith2026250524605,
  author       = {Pith},
  title        = {Pith review of: Model-Guided Network with Cluster-Based Operators for Spatio-Spectral Super-Resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F2NAXYC4}},
  note         = {Machine review of arXiv:2505.24605}
}
read the original abstract

This paper addresses the problem of reconstructing a high-resolution hyperspectral image from a low-resolution multispectral observation. While spatial super-resolution and spectral super-resolution have been extensively studied, joint spatio-spectral super-resolution remains relatively explored. We propose an end-to-end model-driven framework that explicitly decomposes the joint spatio-spectral super-resolution problem into spatial super-resolution, spectral super-resolution and fusion tasks. Each sub-task is addressed by unfolding a variational-based approach, where the operators involved in the proximal gradient iterative scheme are replaced with tailored learnable modules. In particular, we design an upsampling operator for spatial super-resolution based on classical back-projection algorithms, adapted to handle arbitrary scaling factors. Spectral reconstruction is performed using learnable cluster-based upsampling and downsampling operators. For image fusion, we integrate low-frequency estimation and high-frequency injection modules to combine the spatial and spectral information from spatial super-resolution and spectral super-resolution outputs. Additionally, we introduce an efficient nonlocal post-processing step that leverages image self-similarity by combining a multi-head attention mechanism with residual connections. Extensive evaluations on several datasets and sampling factors demonstrate the effectiveness of our approach. The source code will be available at https://github.com/TAMI-UIB/JSSUNet

Figures

Figures reproduced from arXiv: 2505.24605 by the authors.

Figure 1
Figure 1. Overview of the proposed joint spectral and spatial super-resolution ap [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Architectures of two basic blocks used in all modules: the Residual [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Architectures of the proximity operators Prox [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (5 more)
Figure 6
Figure 6. Figure 6: Architectures of the High Frequency Injection (HFI) module (top-left), [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Proposed post-processing architecture (top), Multi-Head Attention [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Visual comparison on CAVE test set for sampling of 2. For each method, the high-resolution result is displayed. Our method achieves sharper contours [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Visual comparison on Pavia test set for sampling of 2. For each method, the high-resolution result is displayed. Our method produces the most vivid [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Visual comparison on NTIRE2020 test set for sampling of 8. For each method, the high-resolution result is displayed. As it can be observed in [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Super-Resolution of Sentinel-2 Images Using a Geometry-Guided Back-Projection Network with Self-Attention

    eess.IV 2025-08 conditional novelty 5.0 of 10

    A geometry-guided, unfolded back-projection network with multi-head self-attention sharpens Sentinel-2's 20m bands to 10m using a cluster-learned guiding image, beating existing fusion methods by about 1 dB PSNR.

Reference graph

Works this paper leans on

72 extracted references · 62 canonical work pages · cited by 1 Pith paper

  1. [1]

    Remote sensing image super-resolution and object detection: Benchmark and state of the art,

    Y . Wang, S. M. A. Bashir, M. Khan, Q. Ullah, R. Wang, Y . Song, Z. Guo, and Y . Niu, “Remote sensing image super-resolution and object detection: Benchmark and state of the art,” Expert Systems with Applications, vol. 197, p. 116793, 2022

    work page 2022

  2. [2]

    Recent advances of hyperspectral imaging technology and applications in agriculture,

    B. Lu, P. D. Dao, J. Liu, Y . He, and J. Shang, “Recent advances of hyperspectral imaging technology and applications in agriculture,” Remote Sensing , vol. 12, no. 16, p. 2659, 2020

    work page 2020

  3. [3]

    Hyperspectral imaging for clinical applications,

    J. Yoon, “Hyperspectral imaging for clinical applications,” BioChip Journal, vol. 16, no. 1, pp. 1–12, 2022

    work page 2022

  4. [4]

    Extraction of spectral information from hyperspectral data and application of hyperspectral imaging for food and agricultural products,

    L. Ravikanth, D. S. Jayas, N. D. White, P. G. Fields, and D.-W. Sun, “Extraction of spectral information from hyperspectral data and application of hyperspectral imaging for food and agricultural products,” F ood and bioprocess technology, vol. 10, pp. 1–33, 2017

    work page 2017

  5. [5]

    A review: Remote sensing sensors,

    L. Zhu, J. Suomalainen, J. Liu, J. Hyypp ¨a, H. Kaartinen, H. Haggren et al., “A review: Remote sensing sensors,” Multi-purposeful application of geospatial data , vol. 19, pp. 19–42, 2018

    work page 2018

  6. [6]

    Hyperspectral image super- resolution using deep convolutional neural network,

    Y . Li, J. Hu, X. Zhao, W. Xie, and J. Li, “Hyperspectral image super- resolution using deep convolutional neural network,” Neurocomputing, vol. 266, pp. 29–41, 2017

    work page 2017

  7. [7]

    Hyperspectral image super-resolution via deep spatiospectral at- tention convolutional neural networks,

    J.-F. Hu, T.-Z. Huang, L.-J. Deng, T.-X. Jiang, G. Vivone, and J. Chanus- sot, “Hyperspectral image super-resolution via deep spatiospectral at- tention convolutional neural networks,” IEEE Transactions on Neural Networks and Learning Systems , vol. 33, no. 12, pp. 7251–7265, 2021

    work page 2021

  8. [8]

    Essaformer: Efficient transformer for hyperspectral image super- resolution,

    M. Zhang, C. Zhang, Q. Zhang, J. Guo, X. Gao, and J. Zhang, “Essaformer: Efficient transformer for hyperspectral image super- resolution,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 23 073–23 084

    work page 2023

Show all 72 references
  1. [9]

    Deep unfolding network for spatiospectral image super-resolution,

    Q. Ma, J. Jiang, X. Liu, and J. Ma, “Deep unfolding network for spatiospectral image super-resolution,” IEEE Transactions on Compu- tational Imaging , vol. 8, pp. 28–40, 2021

    2021

  2. [10]

    Spatial and spectral joint super- resolution using convolutional neural network,

    S. Mei, R. Jiang, X. Li, and Q. Du, “Spatial and spectral joint super- resolution using convolutional neural network,” IEEE Transactions on Geoscience and Remote Sensing , vol. 58, no. 7, pp. 4590–4603, 2020

    2020

  3. [11]

    Multi-task interaction learning for spatiospectral image super-resolution,

    Q. Ma, J. Jiang, X. Liu, and J. Ma, “Multi-task interaction learning for spatiospectral image super-resolution,” IEEE Transactions on Image Processing, vol. 31, pp. 2950–2961, 2022

    2022

  4. [12]

    Motion analysis for image enhancement: Res- olution, occlusion, and transparency,

    M. Irani and S. Peleg, “Motion analysis for image enhancement: Res- olution, occlusion, and transparency,” Journal of visual communication and image representation , vol. 4, no. 4, pp. 324–335, 1993

    1993

  5. [13]

    Robust super-resolution,

    A. Zomet, A. Rav-Acha, and S. Peleg, “Robust super-resolution,” in Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001 , vol. 1. IEEE, 2001, pp. I–I

    2001

  6. [14]

    A survey of pansharpening methods with a new band-decoupled variational model,

    J. Duran, A. Buades, B. Coll, C. Sbert, and G. Blanchet, “A survey of pansharpening methods with a new band-decoupled variational model,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 125, pp. 78–105, 2017

    2017

  7. [15]

    Image super- resolution: A comprehensive review, recent trends, challenges and ap- plications,

    D. C. Lepcha, B. Goyal, A. Dogra, and V . Goyal, “Image super- resolution: A comprehensive review, recent trends, challenges and ap- plications,” Information Fusion , vol. 91, pp. 230–260, 2023

    2023

  8. [16]

    New edge-directed interpolation,

    X. Li and M. T. Orchard, “New edge-directed interpolation,” IEEE transactions on image processing , vol. 10, no. 10, pp. 1521–1527, 2001

    2001

  9. [17]

    Cubic convolution interpolation for digital image processing,

    R. Keys, “Cubic convolution interpolation for digital image processing,” IEEE transactions on acoustics, speech, and signal processing , vol. 29, no. 6, pp. 1153–1160, 1981

    1981

  10. [18]

    Regularization-based multi-frame super-resolution: A systematic review,

    M. Khattab, A. Zeki, A. Alwan, and A. Badawy, “Regularization-based multi-frame super-resolution: A systematic review,” Journal of King Saud University - Computer and Information Sciences , vol. 32, 11 2018

    2018

  11. [19]

    Image deblurring and super- resolution by adaptive sparse domain selection and adaptive regular- ization,

    W. Dong, L. Zhang, G. Shi, and X. Wu, “Image deblurring and super- resolution by adaptive sparse domain selection and adaptive regular- ization,” IEEE Transactions on image processing , vol. 20, no. 7, pp. 1838–1857, 2011

    2011

  12. [20]

    Nonlocally centralized sparse representation for image restoration,

    W. Dong, L. Zhang, G. Shi, and X. Li, “Nonlocally centralized sparse representation for image restoration,” IEEE transactions on Image Processing, vol. 22, no. 4, pp. 1620–1630, 2012

    2012

  13. [21]

    What if image self- similarity can be better exploited in data fidelity terms?

    I. Pereira-S ´anchez, J. Navarro, and J. Duran, “What if image self- similarity can be better exploited in data fidelity terms?” in 2022 IEEE International Conference on Image Processing (ICIP) . IEEE, 2022, pp. 3697–3701

    2022

  14. [22]

    Image super-resolution using deep convolutional networks,

    C. Dong, C. C. Loy, K. He, and X. Tang, “Image super-resolution using deep convolutional networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 38, no. 2, pp. 295–307, 2016

    2016

  15. [23]

    Enhanced deep residual networks for single image super-resolution,

    B. Lim, S. Son, H. Kim, S. Nah, and K. M. Lee, “Enhanced deep residual networks for single image super-resolution,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2017, pp. 1132–1140

    2017

  16. [24]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” in 2017 IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR) , 2017, pp. 2261–2269

    2017

  17. [25]

    A review on single image super resolution techniques using generative adversarial network,

    K. Singla, R. Pandey, and U. Ghanekar, “A review on single image super resolution techniques using generative adversarial network,” Optik, vol. 266, p. 169607, 2022

    2022

  18. [26]

    Dual-domain prior unfolding network for remote sensing image super-resolution,

    J. Dong, G. Hu, J. Zhang, and X. Luo, “Dual-domain prior unfolding network for remote sensing image super-resolution,” Earth Science Informatics, vol. 18, no. 1, p. 128, 2025

    2025

  19. [27]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly et al. , “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

    2010 arXiv

  20. [28]

    Swinir: Image restoration using swin transformer,

    J. Liang, J. Cao, G. Sun, K. Zhang, L. Van Gool, and R. Timofte, “Swinir: Image restoration using swin transformer,” in Proceedings of the IEEE/CVF international conference on computer vision , 2021, pp. 1833–1844

    2021

  21. [29]

    Efficient long-range attention network for image super-resolution,

    X. Zhang, H. Zeng, S. Guo, and L. Zhang, “Efficient long-range attention network for image super-resolution,” in European conference on computer vision . Springer, 2022, pp. 649–667

    2022

  22. [30]

    Hyperspectral image superresolution by transfer learning,

    Y . Yuan, X. Zheng, and X. Lu, “Hyperspectral image superresolution by transfer learning,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , vol. 10, no. 5, pp. 1963–1974, 2017. JOURNAL OF LATEX CLASS FILES, VOL. X, NO. X, MAY 2025 13

    1963

  23. [31]

    Learning spatial-spectral prior for super-resolution of hyperspectral imagery,

    J. Jiang, H. Sun, X. Liu, and J. Ma, “Learning spatial-spectral prior for super-resolution of hyperspectral imagery,” IEEE Transactions on Computational Imaging , vol. 6, pp. 1082–1096, 2020

    2020

  24. [32]

    Spectral super-resolution meets deep learning: Achievements and challenges,

    J. He, Q. Yuan, J. Li, Y . Xiao, D. Liu, H. Shen, and L. Zhang, “Spectral super-resolution meets deep learning: Achievements and challenges,” Inf. Fusion, vol. 97, no. C, Sep. 2023. [Online]. Available: https://doi.org/10.1016/j.inffus.2023.101812

    2023

  25. [33]

    Adversarial networks for spatial context-aware spectral image reconstruction from rgb,

    A. Alvarez-Gila, J. Van De Weijer, and E. Garrote, “Adversarial networks for spatial context-aware spectral image reconstruction from rgb,” in 2017 IEEE International Conference on Computer Vision Workshops (ICCVW) . IEEE, Oct. 2017, p. 480–490. [Online]. Available: http://dx....

    2017 doi

  26. [34]

    Learned spectral super-resolution,

    S. Galliani, C. Lanaras, D. Marmanis, E. Baltsavias, and K. Schindler, “Learned spectral super-resolution,” 2017. [Online]. Available: https: //arxiv.org/abs/1703.09470

    2017 arXiv

  27. [35]

    Accurate spectral super-resolution from single rgb image using multi-scale cnn,

    Y . Yan, L. Zhang, J. Li, W. Wei, and Y . Zhang, “Accurate spectral super-resolution from single rgb image using multi-scale cnn,” 2018. [Online]. Available: https://arxiv.org/abs/1806.03575

    2018 arXiv

  28. [36]

    Adaptive weighted attention network with camera spectral sensitivity prior for spectral reconstruction from rgb images,

    J. Li, C. Wu, R. Song, Y . Li, and F. Liu, “Adaptive weighted attention network with camera spectral sensitivity prior for spectral reconstruction from rgb images,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops , 2020, pp. 462– 463

    2020

  29. [37]

    Rgb to spectral reconstruction via learned basis functions and weights,

    B. J. Fubara, M. Sedky, and D. Dyke, “Rgb to spectral reconstruction via learned basis functions and weights,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , 2020, pp. 1984–1993

    2020

  30. [38]

    Learning hyperspectral images from rgb images via a coarse-to-fine cnn,

    S. Mei, Y . Geng, J. Hou, and Q. Du, “Learning hyperspectral images from rgb images via a coarse-to-fine cnn,”SCIENCE CHINA Information Sciences, vol. 65, no. 5, pp. 152 102–, 2022

    2022

  31. [39]

    Deep amended gradient descent for efficient spectral reconstruction from single rgb images,

    Z. Zhu, H. Liu, J. Hou, S. Jia, and Q. Zhang, “Deep amended gradient descent for efficient spectral reconstruction from single rgb images,” IEEE Transactions on Computational Imaging , vol. 7, pp. 1176–1188, 2021

    2021

  32. [40]

    A new benchmark based on recent advances in multispectral pansharpening: Revisiting pansharpening with classical and emerging pansharpening methods,

    G. Vivone, M. Dalla Mura, A. Garzelli, R. Restaino, G. Scarpa, M. O. Ulfarsson, L. Alparone, and J. Chanussot, “A new benchmark based on recent advances in multispectral pansharpening: Revisiting pansharpening with classical and emerging pansharpening methods,” IEEE Geoscience...

    2020

  33. [41]

    Multispectral and hyperspectral image fusion in remote sensing: A survey,

    G. Vivone, “Multispectral and hyperspectral image fusion in remote sensing: A survey,” Information Fusion , vol. 89, pp. 405–417, 2023. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S1566253522001312

    2023

  34. [42]

    A comprehensive overview of satellite image fusion: From classical model-based to cutting-edge deep learning approaches,

    I. Pereira-S ´anchez, E. Sans, J. Navarro, and J. Duran, “A comprehensive overview of satellite image fusion: From classical model-based to cutting-edge deep learning approaches,” Super-Resolution for Remote Sensing, pp. 279–328, 2024

    2024

  35. [43]

    Color enhancement of highly correlated images. ii. channel ratio and “chromaticity

    A. R. Gillespie, A. B. Kahle, and R. E. Walker, “Color enhancement of highly correlated images. ii. channel ratio and “chromaticity” transfor- mation techniques,” Remote Sensing of Environment , vol. 22, no. 3, pp. 343–365, 1987

    1987

  36. [44]

    Comparison of three different methods to merge multiresolution and multispectral data- landsat tm and spot panchromatic,

    P. Chavez, S. C. Sides, J. A. Anderson et al. , “Comparison of three different methods to merge multiresolution and multispectral data- landsat tm and spot panchromatic,” Photogrammetric Engineering and remote sensing, vol. 57, no. 3, pp. 295–303, 1991

    1991

  37. [45]

    Improving component substitution pansharpening through multivariate regression of ms + pan data,

    B. Aiazzi, S. Baronti, and M. Selva, “Improving component substitution pansharpening through multivariate regression of ms + pan data,” IEEE Transactions on Geoscience and Remote Sensing , vol. 45, no. 10, pp. 3230–3239, 2007

    2007

  38. [46]

    Mtf- tailored multiscale fusion of high-resolution ms and pan imagery,

    B. Aiazzi, L. Alparone, S. Baronti, A. Garzelli, and M. Selva, “Mtf- tailored multiscale fusion of high-resolution ms and pan imagery,” Photogrammetric Engineering & Remote Sensing , vol. 72, no. 5, pp. 591–596, 2006

    2006

  39. [47]

    A wavelet based algorithm for pan sharpening landsat 7 imagery,

    R. L. King and J. Wang, “A wavelet based algorithm for pan sharpening landsat 7 imagery,” inIGARSS 2001. Scanning the Present and Resolving the Future. Proceedings. IEEE 2001 International Geoscience and Remote Sensing Symposium (Cat. No. 01CH37217) , vol. 2. IEEE, 2001, pp. 849–851

    2001

  40. [48]

    Introduction of sensor spectral response into image fusion methods. application to wavelet-based methods,

    X. Otazu, M. Gonz ´alez-Aud´ıcana, O. Fors, and J. N ´u˜nez, “Introduction of sensor spectral response into image fusion methods. application to wavelet-based methods,” IEEE Transactions on Geoscience and Remote Sensing, vol. 43, no. 10, pp. 2376–2385, 2005

    2005

  41. [49]

    Fusion of multispectral and panchromatic images by local mean and variance matching filtering techniques,

    S. de B ´ethune, F. Muller, and J.-P. Donnay, “Fusion of multispectral and panchromatic images by local mean and variance matching filtering techniques,” Fusion of Earth Data , vol. 28, p. 30, 1998

    1998

  42. [50]

    Pansharpening quality assessment using the modulation transfer functions of instruments,

    M. M. Khan, L. Alparone, and J. Chanussot, “Pansharpening quality assessment using the modulation transfer functions of instruments,” IEEE Transactions on Geoscience and Remote Sensing , vol. 47, no. 11, pp. 3880–3891, 2009

    2009

  43. [51]

    A variational model for p+ xs image fusion,

    C. Ballester, V . Caselles, L. Igual, J. Verdera, and B. Roug ´e, “A variational model for p+ xs image fusion,” International Journal of Computer Vision, vol. 69, pp. 43–58, 2006

    2006

  44. [52]

    A nonlocal variational model for pansharpening image fusion,

    J. Duran, A. Buades, B. Coll, and C. Sbert, “A nonlocal variational model for pansharpening image fusion,” SIAM Journal on Imaging Sciences , vol. 7, no. 2, pp. 761–796, 2014

    2014

  45. [53]

    Pansharpening via double nonconvex tensor low- tubal-rank priors,

    P. Liu and Z. Zheng, “Pansharpening via double nonconvex tensor low- tubal-rank priors,” IEEE Geoscience and Remote Sensing Letters , 2024

    2024

  46. [54]

    Pannet: A deep network architecture for pan-sharpening,

    J. Yang, X. Fu, Y . Hu, Y . Huang, X. Ding, and J. Paisley, “Pannet: A deep network architecture for pan-sharpening,” in Proceedings of the IEEE International Conference on Computer Vision (ICCV) , Oct 2017

    2017

  47. [55]

    Pansharp- ening via detail injection based convolutional neural networks,

    L. He, Y . Rao, J. Li, J. Chanussot, A. Plaza, J. Zhu, and B. Li, “Pansharp- ening via detail injection based convolutional neural networks,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 12, no. 4, pp. 1188–1204, 2019

    2019

  48. [56]

    Super-resolution-guided progressive pansharpen- ing based on a deep convolutional neural network,

    J. Cai and B. Huang, “Super-resolution-guided progressive pansharpen- ing based on a deep convolutional neural network,” IEEE Transactions on Geoscience and Remote Sensing , vol. 59, no. 6, pp. 5206–5220, 2020

    2020

  49. [57]

    Awfln: An adaptive weighted feature learning network for pansharpening,

    H. Lu, Y . Yang, S. Huang, X. Chen, B. Chi, A. Liu, and W. Tu, “Awfln: An adaptive weighted feature learning network for pansharpening,”IEEE Transactions on Geoscience and Remote Sensing , vol. 61, pp. 1–15, 2023

    2023

  50. [58]

    Hypertransformer: A textural and spectral feature fusion transformer for pansharpening,

    W. G. C. Bandara and V . M. Patel, “Hypertransformer: A textural and spectral feature fusion transformer for pansharpening,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2022, pp. 1767–1777

    2022

  51. [59]

    Panformer: A transformer based model for pan-sharpening,

    H. Zhou, Q. Liu, and Y . Wang, “Panformer: A transformer based model for pan-sharpening,” in 2022 IEEE International Conference on Multimedia and Expo (ICME) . IEEE, 2022, pp. 1–6

    2022

  52. [60]

    Fusformer: A transformer-based fusion network for hyperspectral im- age super-resolution,

    J.-F. Hu, T.-Z. Huang, L.-J. Deng, H.-X. Dou, D. Hong, and G. Vivone, “Fusformer: A transformer-based fusion network for hyperspectral im- age super-resolution,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

    2022

  53. [61]

    Deep unfolding tensor rank minimization with generalized detail injection for pansharpening,

    T. T. N. Mai, E. Y . Lam, and C. Lee, “Deep unfolding tensor rank minimization with generalized detail injection for pansharpening,” IEEE Transactions on Geoscience and Remote Sensing , 2024

    2024

  54. [62]

    Spatial-spectral dual back-projection network for pansharpening,

    K. Zhang, A. Wang, F. Zhang, W. Wan, J. Sun, and L. Bruzzone, “Spatial-spectral dual back-projection network for pansharpening,” IEEE Transactions on Geoscience and Remote Sensing , 2023

    2023

  55. [63]

    Distributed optimization and statistical learning via the alternating direction method of multipliers,

    S. Boyd, N. Parikh, E. Chu, B. Peleato, J. Eckstein et al. , “Distributed optimization and statistical learning via the alternating direction method of multipliers,” F oundations and Trends® in Machine learning , vol. 3, no. 1, pp. 1–122, 2011

    2011

  56. [64]

    Hyperspectral image joint super- resolution via local implicit spatial-spectral function learning,

    Y . Zhang, J. Zhang, and S. Han, “Hyperspectral image joint super- resolution via local implicit spatial-spectral function learning,” IEEE Photonics Journal, 2024

    2024

  57. [65]

    Proximal algorithms,

    N. Parikh, S. Boyd et al. , “Proximal algorithms,” F oundations and trends® in Optimization , vol. 1, no. 3, pp. 127–239, 2014

    2014

  58. [66]

    An introduction to continuous optimization for imaging,

    A. Chambolle and T. Pock, “An introduction to continuous optimization for imaging,” Acta Numerica , vol. 25, pp. 161–319, 2016

    2016

  59. [67]

    Image super- resolution using very deep residual channel attention networks,

    Y . Zhang, K. Li, K. Li, L. Wang, B. Zhong, and Y . Fu, “Image super- resolution using very deep residual channel attention networks,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 286–301

    2018

  60. [68]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

  61. [69]

    Generalized assorted pixel camera: postcapture control of resolution, dynamic range, and spectrum,

    F. Yasuma, T. Mitsunaga, D. Iso, and S. K. Nayar, “Generalized assorted pixel camera: postcapture control of resolution, dynamic range, and spectrum,” IEEE transactions on image processing , vol. 19, no. 9, pp. 2241–2253, 2010

    2010

  62. [70]

    Ntire 2020 challenge on spectral reconstruction from an rgb image,

    B. Arad, R. Timofte, O. Ben-Shahar, Y .-T. Lin, and G. D. Finlayson, “Ntire 2020 challenge on spectral reconstruction from an rgb image,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops , 2020, pp. 446–447

    2020

  63. [71]

    Rabbani and P

    M. Rabbani and P. W. Jones, Digital image compression techniques . SPIE press, 1991, vol. 7

    1991

  64. [72]

    A universal image quality index,

    Z. Wang and A. Bovik, “A universal image quality index,” IEEE Signal Processing Letters, vol. 9, no. 3, pp. 81–84, 2002

    2002

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.