Pith. sign in

REVIEW 3 major objections 5 minor 70 references

Clarity Contrast and Similarity Selection for Multi-Focus Image Fusion

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read CSNet claims that multi-focus image fusion improves when the two source images explicitly contrast their clarity at multiple scales and ambiguous boundary pixels are chosen by similarity to a reconstructed sharp image.

desk verdict A genuinely new interaction idea for multi-focus fusion, but the SOTA claim is propped up by test-set hyperparameter tuning and single-run metrics. read the letter →

arxiv 2608.03252 v1 pith:3UR4AX3P submitted 2026-08-04 cs.CV

Yicheng Zhang , Haoyou Deng , Zhiqiang Li , Wenti Yin , Nong Sang , Changxin Gao This is my paper
classification cs.CV
keywords multi-focusimagefusionclaritycontrastattentionfocusmapboundaryrefinementsimilarityselectiondefocusspreadeffectdeeplearning
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

Multi-focus image fusion combines two photos of the same scene taken with different focal planes into one all-in-focus image. Most deep-learning fusion networks feed the two sources into a single black-box network without letting them compare notes, which hurts both accuracy and interpretability near focused-defocused boundaries. The paper proposes CSNet, a two-branch network in which the sources explicitly contrast their clarity at multiple scales: each branch uses the other's strength to suppress its own blurry regions, producing focus maps that mark exactly which pixels to take from which source. A separate similarity selection step reconstructs an initial sharp image, then decides boundary pixels by comparing perceptual similarity between the reconstruction and each source, choosing the most faithful pixel or the reconstruction itself. The paper reports that this interactive design outperforms previous methods quantitatively and qualitatively on four standard benchmarks, with most gains concentrated at boundaries where both sources are degraded by defocus spread.

What carries the argument

The load-bearing mechanism is the inverse clarity modulation inside CCAM, together with the similarity thresholding at the boundary. A clarity map encodes where each source is sharper; the difference between the two maps acts as a spatial weight that suppresses blurry content in the other branch. This produces binary focus maps and a residual boundary map. Then the Similarity Selection Strategy computes perceptual distance between source features and reconstructed features using a pretrained VGG-16 network, and Eq. (12) chooses source pixels when similarity is strongly biased, otherwise uses the reconstruction. The identity that carries the argument is $I_F = I_A \odot M_A + I_B \odot M_B +

What would settle it

Run CSNet on a real or synthetic pair where the focus difference is deliberately made small (e.g., Gaussian blur with $\sigma$ near the low end of the training range) but the true focus mask is known from a depth map or manual annotation. If the predicted boundary map leaks into clearly focused regions, or the focus maps swap labels on low-texture defocused areas, the inverse-modulation assumption is violated. The paper's failure cases in Figure 10, where 'minor inaccuracies' appear under subtle blur and manual annotation, point to exactly this test.

Watch

Extended reading notes

Core claim

The central claim is that the two source images should be made to interact explicitly through a Clarity Contrast Attention Module (CCAM). At each encoder level, the network produces clarity maps $C_A$ and $C_B$ for the two sources, forms the difference $D_A = C_A - C_B$ (and symmetrically $D_B$), and uses it to inversely modulate the other branch: $Y'_A = Y_A \odot (1 - D_B)$ and $Y'_B = Y_B \odot (1 - D_A)$. This contrastive modulation mutually sharpens focused features and suppresses blurred ones, yielding focus maps $M_A$ and $M_B$ and a boundary map $M_C = 1 - M_A - M_B$. For the ambiguous boundary band, a Similarity Selection Strategy reconstructs an initial clear image $I_C$ via cross-

Load-bearing premise

The whole pipeline rests on the learned clarity maps—trained on synthetic Gaussian blur—ranking true focus correctly on real photographs, so that suppressing the branch judged 'less clear' removes exactly the defocused content and the residual boundary map captures the real defocus-spread band.

Editorial extensions

If this is right

  • If CSNet's design is correct, explicit inter-source interaction is sufficient to localize focused regions more accurately than the common single-stream or concatenation-based networks.
  • Boundary artifacts caused by the Defocus Spread Effect can be reduced by reconstructing an initial sharp image and selecting boundary pixels by perceptual similarity rather than by forcing a hard decision between sources.
  • The same architecture should extend to more than two input images by processing them in a staged pairwise manner, as the paper states.
  • The trained focus maps carry a clear interpretation: they are binary masks of focused regions plus a residual boundary band, which makes the fusion process inspectable and potentially useful for downstream tasks that need focus segmentation.

Reading between the lines

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

  • Outside the paper: the same clarity-contrast idea could transfer to other fusion- or restoration settings where two degraded observations of the same scene are available, such as infrared-visible fusion or stereo super-resolution, because the contrastive suppression does not depend on the specific defocus blur model.
  • The fixed threshold $\theta=1.2$ is a hand-set balance between fidelity and smoothness; a reader could test whether learning the threshold per pixel, or conditioning it on local blur estimates, improves boundary fidelity further.
  • Since the network is trained on synthetic Gaussian blur, a plausible stress test is to apply it to images with non-Gaussian or spatially varying blur; the paper's own failure cases already suggest boundary leakage when blur is subtle or annotations are manual, so the claimed advantage may shrink on very different blur kernels.
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 CSNet, a deep network for multi-focus image fusion that explicitly models inter-source interaction. The Clarity Contrast Attention Module (CCAM) contrasts clarity maps from the two source images and inversely modulates their features to generate focus maps and a boundary map. The Similarity Selection Strategy (SSS) reconstructs an initial clear image via cross-attention and refines boundary pixels by comparing perceptual similarities among sources and the reconstruction. The network is trained on synthetic pairs with a simulated Defocus Spread Effect and evaluated on Lytro, MFFW, MFI-WHU, and SIMIF against 15 methods. The paper claims state-of-the-art quantitative and qualitative performance, and releases code.

Significance. If the empirical results are robust, the paper makes a useful contribution: the explicit clarity-contrast interaction is interpretable, and the boundary refinement via perceptual similarity is a sensible treatment of the defocus spread effect. The ablations are extensive and largely validate the design choices, and the public code aids reproducibility. However, the central claim of state-of-the-art performance rests on an evaluation protocol that is not statistically secured: hyperparameters are selected on the test benchmarks, reported margins over prior methods are often in the third or fourth decimal, and no variance or significance analysis is given. The contribution is therefore plausible but not yet demonstrated to the standard required for a strong performance claim.

major comments (3)
  1. [§IV-D, Table IV] The hyperparameters θ and (λ1, λ2, μ) are selected by evaluating on the MFFW and Lytro test benchmarks. The sensitivity table shows that the metric variation across θ is of the same order as the reported advantage over the runner-up: e.g., MFFW QMI is 1.1842 at θ=1.15 and 1.1858 at θ=1.20, while the best competitor (MCCSR) achieves 1.1768; on Lytro, QMI ranges 1.1905–1.1924 across θ, versus 1.1920 for MCCSR. Thus the selected θ could account for the claimed margin. The paper must use a held-out validation split or cross-validation for hyperparameter selection and report test-set results after freezing all choices. In addition, no error bars, multiple seeds, or statistical tests are reported; all numbers appear to come from a single run. This is load-bearing for the 'state-of-the-art' claim.
  2. [§IV-A, Tables I–II] The comparison protocol is asymmetric: all competitor methods are evaluated with their official pre-trained weights, while CSNet's hyperparameters are tuned on the same test benchmarks. This gives CSNet an advantage that is particularly consequential given the small metric gaps. The authors should either (a) select hyperparameters on a validation set and only then evaluate on the test sets, or (b) if test-set tuning is unavoidable, clearly label the results as best-case selected values and temper the abstract's claim. Without such a change, the evidence does not support 'state-of-the-art performance' as stated.
  3. [§III-D and §IV-A] The similarity threshold θ (Eq. 12) and the 'small region removal' threshold (0.01×H×W) are both user-set parameters that directly affect the fused output. Only θ is analyzed, and its optimal value is test-selected. The small-region threshold is not varied at all. Please provide a sensitivity analysis for this parameter as well, or at least justify the choice independently of test-set performance. This is part of the broader issue that several free parameters are fixed based on test benchmarks, weakening the internal validity of the reported gains.
minor comments (5)
  1. [Abstract and §IV-B] The abstract claims 'state-of-the-art performance both quantitatively and qualitatively,' but on Lytro the method does not achieve the best score on QG, QY, or QC (Table I: e.g., QG 0.7592 vs MCCSR 0.7619). Recommend wording such as 'competitive or state-of-the-art on most metrics.'
  2. [§III-D, Eq. (11)] The similarity measure S_A and S_B is defined as the inverse of a normalized Euclidean distance. Please specify which VGG-16 layers are used for 'multi-scale features' and clarify that a larger S value corresponds to higher similarity. The threshold comparison S_A > θ·S_B is then intuitive, but the inverse relationship should be stated explicitly.
  3. [§III-B] Typographical issues: 'an spatial characterization' and 'the the resulting feature' appear in the text. Also in §IV-C, 'Lyrto' is a misspelling of 'Lytro.'
  4. [§IV-A] The implementation details state that training uses 50 epochs and a single RTX 4090, but the number of random seeds or repeated runs is not reported. Please state that each result is a single run, or provide variance information.
  5. [§IV-A, Table II] The time measurement for CSNet is 0.062 s; please clarify whether this includes the VGG-based similarity computation and the post-processing small-region removal, or only the network forward pass.

Circularity Check

1 steps flagged · score 5.0 of 10

Test-set hyperparameter selection underlies the reported SOTA; architecture derivation is otherwise independent.

  1. fitted input called prediction [Section IV-D (Sensitivity Analysis), Table IV; applied to Tables I and II]
    "while θ=1.20 achieves the best performance on most metrics... The setting (λ1, λ2, µ) = (1.0,0.5,0.5) achieves the strongest performance."

    The similarity threshold θ and loss weights (λ1,λ2,µ) are selected by inspecting the six quantitative metrics on the MFFW and Lytro test benchmarks in Table IV. The final reported results in Tables I and II, which support the abstract's 'state-of-the-art' claim, are produced with these selected values. Thus the superiority over prior methods is not an independent prediction of a fixed model; it is the outcome of a search over the evaluation function. The specific margins (e.g., MFFW QMI 1.1858 vs. 1.1768) reflect the optimization of θ and loss weights, not a pre-specified configuration. This is a fitted parameter called a prediction.

full rationale

The paper's method (CCAM, cross-reconstruction, similarity selection, Eqs. 1-16) is a coherent, self-contained architecture with no definitional circularity. The synthetic training data and VGG-based similarity are external evidence. The only circular element is the empirical claim: θ and loss weights are chosen by evaluating the exact metrics on the exact test sets later used to demonstrate SOTA (Section IV-D, Table IV). Since the metric gaps are in the third/fourth decimal and comparable to variation across θ values, the reported advantages are not independent of the selection. Nevertheless, the architecture's components are supported by ablations, and results on MFI-WHU and SIMIF provide some independent evidence. Therefore the paper has partial circularity in the evaluation, not in the derivation.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The method relies on the physical assumption of complementary focus and on learned or pretrained representations generalizing from synthetic to real images. The main free parameters are tuning choices, including two selected directly on test benchmarks. No new physical entities are introduced.

free parameters (3)
  • similarity threshold theta = 1.20
    Threshold in Eq. (12) for selecting source vs reconstructed pixels. Chosen via sensitivity analysis on MFFW and Lytro test datasets (Table IV), which is a form of test-set tuning.
  • loss weights lambda1, lambda2, mu = 1.0, 0.5, 0.5
    Weights in Eq. (16) balancing focus map loss, reconstruction loss, and SSIM term. Selected using test benchmark performance in Table IV, again test-set tuning.
  • small region removal threshold = 0.01 * H * W
    Post-processing threshold to remove small regions from decision maps, taken from prior work (reference [54]). Affects final focus maps.
assumptions (4)
  • domain assumption In non-boundary areas, the two source images have complementary focus, so clarity is negatively correlated between them.
    Stated in the introduction and used to justify the contrastive modulation in Section III-B. If both images are blurred or both sharp in the same region, the contrast mechanism loses its grounding.
  • domain assumption The learned spatial-attention clarity maps C_A and C_B reliably reflect true focus quality on real images.
    The CCAM relies on these maps to compute differences D_A and D_B in Eq. (3). They are trained on synthetic data and the paper's failure cases (Figure 10) show this assumption breaks in some real scenarios.
  • domain assumption Synthetic data generated with Gaussian blur and blurred masks adequately emulates the real Defocus Spread Effect.
    Section III-F describes training data generation. The claim that the method improves real DSE handling depends on this transfer. The paper's own test-set results on MFFW support it, but the assumption remains unverified for extreme defocus.
  • domain assumption VGG-16 perceptual features are a suitable measure of similarity between source pixels and the reconstructed image.
    Section III-D uses Eq. (11) with pretrained VGG-16 features to drive pixel selection. This assumes perceptual distance corresponds to the source contribution in the reconstructed image.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Clarity Contrast and Similarity Selection for Multi-Focus Image Fusion." pith.science (2026). https://pith.science/paper/3UR4AX3P

@misc{pith2026260803252,
  author       = {Pith},
  title        = {Pith review of: Clarity Contrast and Similarity Selection for Multi-Focus Image Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3UR4AX3P}},
  note         = {Machine review of arXiv:2608.03252}
}
read the original abstract

Multi-focus image fusion (MFIF) aims to generate an all-in-focus image from multiple images of the same scene focused at different regions. Most existing deep learning-based methods lack explicit interaction between the source images, which limits their performance and interpretability. This paper presents a novel Clarity Contrast and Similarity Selection Network (CSNet), to bridge direct information exchange for MFIF. Specifically, by contrasting the clarity differences between source images within our proposed Clarity Contrast Attention Module (CCAM), we mutually enhance sharp features while suppressing blurry ones. This allows us to identify the exactly focused regions in each source and locate the focused-defocused boundaries. Moreover, the Defocus Spread Effect (DSE) degrades pixels in all source images around the boundaries. To further refine these ambiguous areas, we introduce a Similarity Selection Strategy, which reconstructs an initial clear image from source images and selects optimal pixels by comparing the similarity among them. Through this interactive approach, CSNet effectively preserves focused regions as well as recovering natural boundaries to fuse an all-in-focus output. Extensive experiments demonstrate that our method achieves state-of-the-art performance both quantitatively and qualitatively. Our code is available on Github: https://github.com/ZYC-HUST/CSNet.

Figures

Figures reproduced from arXiv: 2608.03252 by the authors.

Figure 1
Figure 1. Illustration of our motivation. (a) Most of the previous works process [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An example of MFIF task and our method. The first row shows source [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The architecture of our CSNet. Given the source image pair, siamese-style encoders are first utilized to extract clarity features. Subsequently, we [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Illustration of our Similarity Selection Strategy. The focus maps identify the focused regions in the source images, which should be directly preserved. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Details of the channel attention module and spatial attention module. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Visualization of the focus maps and boundary maps generated by our CSNet. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Visualization of the focus maps on the Lytro and MFFW datasets. The white regions indicate the focused pixels in the source image [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the fusion images from the Lytro, MFFW and MFI-WHU datasets. Two enlarged views are shown to reveal critical details. [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: Visualization of the fusion images from the Lytro, MFFW and MFI-WHU datasets. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: Representative failure cases of the proposed method. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

70 extracted references · 62 canonical work pages

  1. [1]

    Algebraic multi-grid based multi- focus image fusion using watershed algorithm,

    Y . Huang, W. Li, M. Gao, and Z. Liu, “Algebraic multi-grid based multi- focus image fusion using watershed algorithm,”IEEE Access, vol. 6, pp. 47 082–47 091, 2018

    work page 2018

  2. [2]

    Conditional random field model for robust multi-focus image fusion,

    O. Bouzos, I. Andreadis, and N. Mitianoudis, “Conditional random field model for robust multi-focus image fusion,”IEEE Transactions on Image Processing, vol. 28, no. 11, pp. 5636–5648, 2019

    work page 2019

  3. [3]

    Image fusion with guided filtering,

    S. Li, X. Kang, and J. Hu, “Image fusion with guided filtering,”IEEE Transactions on Image processing, vol. 22, no. 7, pp. 2864–2875, 2013

    work page 2013

  4. [4]

    Image fusion with convolutional sparse representation,

    Y . Liu, X. Chen, R. K. Ward, and Z. J. Wang, “Image fusion with convolutional sparse representation,”IEEE signal processing letters, vol. 23, no. 12, pp. 1882–1886, 2016

    work page 2016

  5. [5]

    Depth-distilled multi-focus image fusion,

    F. Zhao, W. Zhao, H. Lu, Y . Liu, L. Yao, and Y . Liu, “Depth-distilled multi-focus image fusion,”IEEE Transactions on Multimedia, vol. 25, pp. 966–978, 2021

    work page 2021

  6. [6]

    Multi-focus image fusion based on multi-scale gradients and image matting,

    J. Chen, X. Li, L. Luo, and J. Ma, “Multi-focus image fusion based on multi-scale gradients and image matting,”IEEE Transactions on Multimedia, vol. 24, pp. 655–667, 2021

    work page 2021

  7. [7]

    Mlnet: A multi-domain lightweight net- work for multi-focus image fusion,

    X. Nie, B. Hu, and X. Gao, “Mlnet: A multi-domain lightweight net- work for multi-focus image fusion,”IEEE Transactions on Multimedia, vol. 25, pp. 5565–5579, 2022

    work page 2022

  8. [8]

    Densefuse: A fusion approach to infrared and visible images,

    H. Li and X.-J. Wu, “Densefuse: A fusion approach to infrared and visible images,”IEEE Transactions on Image Processing, vol. 28, no. 5, pp. 2614–2623, 2018

    work page 2018

Show all 70 references
  1. [9]

    Dual-path deep unsupervised learning for multi-focus image fusion,

    Y . Quan, X. Wan, T. Zheng, Y . Huang, and H. Ji, “Dual-path deep unsupervised learning for multi-focus image fusion,”IEEE Transactions on Multimedia, 2024

    2024

  2. [10]

    Model- informed multistage unsupervised network for hyperspectral image super-resolution,

    J. Li, K. Zheng, L. Gao, L. Ni, M. Huang, and J. Chanussot, “Model- informed multistage unsupervised network for hyperspectral image super-resolution,”IEEE Transactions on Geoscience and Remote Sens- ing, vol. 62, pp. 1–17, 2024

    2024

  3. [11]

    Multi-focus image fusion: A survey of the state of the art,

    Y . Liu, L. Wang, J. Cheng, C. Li, and X. Chen, “Multi-focus image fusion: A survey of the state of the art,”Information Fusion, vol. 64, pp. 71–91, 2020

    2020

  4. [12]

    Image matting for fusion of multi- focus images in dynamic scenes,

    S. Li, X. Kang, J. Hu, and B. Yang, “Image matting for fusion of multi- focus images in dynamic scenes,”Information Fusion, vol. 14, no. 2, pp. 147–162, 2013

    2013

  5. [13]

    High quality multi-focus image fusion using self-similarity and depth information,

    D. Guo, J. Yan, and X. Qu, “High quality multi-focus image fusion using self-similarity and depth information,”Optics communications, vol. 338, pp. 138–144, 2015

    2015

  6. [14]

    A region-based multi-sensor image fusion scheme using pulse-coupled neural network,

    M. Li, W. Cai, and Z. Tan, “A region-based multi-sensor image fusion scheme using pulse-coupled neural network,”Pattern Recognition Let- ters, vol. 27, no. 16, pp. 1948–1956, 2006

    1948

  7. [15]

    Robust sparse rep- resentation based multi-focus image fusion with dictionary construction and local spatial consistency,

    Q. Zhang, T. Shi, F. Wang, R. S. Blum, and J. Han, “Robust sparse rep- resentation based multi-focus image fusion with dictionary construction and local spatial consistency,”Pattern Recognition, vol. 83, pp. 299–313, 2018

    2018

  8. [16]

    A multi-focus image fusion method via region mosaicking on laplacian pyramids,

    L. Kou, L. Zhang, K. Zhang, J. Sun, Q. Han, and Z. Jin, “A multi-focus image fusion method via region mosaicking on laplacian pyramids,” PloS one, vol. 13, no. 5, p. e0191085, 2018

    2018

  9. [17]

    Multi-scale weighted gradient-based fusion for multi-focus images,

    Z. Zhou, S. Li, and B. Wang, “Multi-scale weighted gradient-based fusion for multi-focus images,”Information Fusion, vol. 20, pp. 60–72, 2014

    2014

  10. [18]

    Deep learning-based multi-focus image fusion: A survey and a comparative study,

    X. Zhang, “Deep learning-based multi-focus image fusion: A survey and a comparative study,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 9, pp. 4819–4838, 2021

    2021

  11. [19]

    Multi-focus image fusion with a deep convolutional neural network,

    Y . Liu, X. Chen, H. Peng, and Z. Wang, “Multi-focus image fusion with a deep convolutional neural network,”Information Fusion, vol. 36, pp. 191–207, 2017

    2017

  12. [20]

    Multilevel features convo- lutional neural network for multifocus image fusion,

    Y . Yang, Z. Nie, S. Huang, P. Lin, and J. Wu, “Multilevel features convo- lutional neural network for multifocus image fusion,”IEEE Transactions on Computational Imaging, vol. 5, no. 2, pp. 262–273, 2019

    2019

  13. [21]

    Global-feature encoding u-net (geu-net) for multi-focus image fusion,

    B. Xiao, B. Xu, X. Bi, and W. Li, “Global-feature encoding u-net (geu-net) for multi-focus image fusion,”IEEE Transactions on Image Processing, vol. 30, pp. 163–175, 2020

    2020

  14. [22]

    Multi-scale visual attention deep convolutional neural network for multi-focus image fusion,

    R. Lai, Y . Li, J. Guan, and A. Xiong, “Multi-scale visual attention deep convolutional neural network for multi-focus image fusion,”IEEE Access, vol. 7, pp. 114 385–114 399, 2019

    2019

  15. [23]

    When multi-focus image fusion networks meet traditional edge- preservation technology,

    Z. Wang, X. Li, L. Zhao, H. Duan, S. Wang, H. Liu, and X. Zhang, “When multi-focus image fusion networks meet traditional edge- preservation technology,”International Journal of Computer Vision, vol. 131, no. 10, pp. 2529–2552, 2023

    2023

  16. [24]

    Ifcnn: A general image fusion framework based on convolutional neural network,

    Y . Zhang, Y . Liu, P. Sun, H. Yan, X. Zhao, and L. Zhang, “Ifcnn: A general image fusion framework based on convolutional neural network,” Information Fusion, vol. 54, pp. 99–118, 2020. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14

    2020

  17. [25]

    Multi-focus image fusion using fully convolutional two-stream network for visual sensors

    K. Xu, Z. Qin, G. Wang, H. Zhang, K. Huang, and S. Ye, “Multi-focus image fusion using fully convolutional two-stream network for visual sensors.”KSII Transactions on Internet & Information Systems, vol. 12, no. 5, 2018

    2018

  18. [26]

    Swinfusion: Cross-domain long-range learning for general image fusion via swin transformer,

    J. Ma, L. Tang, F. Fan, J. Huang, X. Mei, and Y . Ma, “Swinfusion: Cross-domain long-range learning for general image fusion via swin transformer,”IEEE/CAA Journal of Automatica Sinica, vol. 9, no. 7, pp. 1200–1217, 2022

    2022

  19. [27]

    Efficient perceiving local details via adaptive spatial- frequency information integration for multi-focus image fusion,

    J. Huang, J. Tu, G. Meng, Y . Wang, Y . Dong, X. Tu, X. Ding, and Y . Huang, “Efficient perceiving local details via adaptive spatial- frequency information integration for multi-focus image fusion,” in Proceedings of the 32nd ACM International Conference on Multimedia, 2024, ...

    2024

  20. [28]

    Fusion2void: Unsupervised multi-focus image fusion based on image inpainting,

    H. Lin, Y . Lin, J. Xia, L. Fan, F. Li, Y . Wang, and X. Ding, “Fusion2void: Unsupervised multi-focus image fusion based on image inpainting,” IEEE Transactions on Circuits and Systems for Video Technology, 2024

    2024

  21. [29]

    Multi-focus image fusion with deep residual learning and focus property detection,

    Y . Liu, L. Wang, H. Li, and X. Chen, “Multi-focus image fusion with deep residual learning and focus property detection,”Information Fusion, vol. 86, pp. 1–16, 2022

    2022

  22. [30]

    Generation and recombination for multifocus image fusion with free number of inputs,

    H. Li, D. Wang, Y . Huang, Y . Zhang, and Z. Yu, “Generation and recombination for multifocus image fusion with free number of inputs,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 7, pp. 6009–6023, 2023

    2023

  23. [31]

    Exploit the best of both end-to-end and map-based methods for multi-focus image fusion,

    J. Zhang, Q. Liao, H. Ma, J.-H. Xue, W. Yang, and S. Liu, “Exploit the best of both end-to-end and map-based methods for multi-focus image fusion,”IEEE Transactions on Multimedia, vol. 26, pp. 6411– 6423, 2024

    2024

  24. [32]

    Multi-focus image fusion via explicit defocus blur modelling,

    Y . Quan, X. Wan, Z. Tang, J. Liang, and H. Ji, “Multi-focus image fusion via explicit defocus blur modelling,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 6, 2025, pp. 6657– 6665

    2025

  25. [33]

    Cross aggregation transformer for image restoration,

    Z. Chen, Y . Zhang, J. Gu, L. Kong, X. Yuanet al., “Cross aggregation transformer for image restoration,”Advances in Neural Information Processing Systems, vol. 35, pp. 25 478–25 490, 2022

    2022

  26. [34]

    Activating more pixels in image super-resolution transformer,

    X. Chen, X. Wang, J. Zhou, Y . Qiao, and C. Dong, “Activating more pixels in image super-resolution transformer,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 22 367–22 377

    2023

  27. [35]

    Text-if: Leveraging semantic text guidance for degradation-aware and interactive image fusion,

    X. Yi, H. Xu, H. Zhang, L. Tang, and J. Ma, “Text-if: Leveraging semantic text guidance for degradation-aware and interactive image fusion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 27 026–27 035

    2024

  28. [36]

    Cddfuse: Correlation-driven dual-branch feature decomposition for multi-modality image fusion,

    Z. Zhao, H. Bai, J. Zhang, Y . Zhang, S. Xu, Z. Lin, R. Timofte, and L. Van Gool, “Cddfuse: Correlation-driven dual-branch feature decomposition for multi-modality image fusion,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 5906–5916

    2023

  29. [37]

    Disparity-aware domain adaptation in stereo image restoration,

    B. Yan, C. Ma, B. Bare, W. Tan, and S. C. Hoi, “Disparity-aware domain adaptation in stereo image restoration,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 13 179–13 187

    2020

  30. [38]

    Steformer: Efficient stereo image super- resolution with transformer,

    J. Lin, L. Yin, and Y . Wang, “Steformer: Efficient stereo image super- resolution with transformer,”IEEE Transactions on Multimedia, vol. 25, pp. 8396–8407, 2023

    2023

  31. [39]

    Drpl: Deep regression pair learning for multi-focus image fusion,

    J. Li, X. Guo, G. Lu, B. Zhang, Y . Xu, F. Wu, and D. Zhang, “Drpl: Deep regression pair learning for multi-focus image fusion,”IEEE Transactions on Image Processing, vol. 29, pp. 4816–4831, 2020

    2020

  32. [40]

    Dtmnet: A discrete tchebichef moments- based deep neural network for multi-focus image fusion,

    B. Xiao, H. Wu, and X. Bi, “Dtmnet: A discrete tchebichef moments- based deep neural network for multi-focus image fusion,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 43–51

    2021

  33. [41]

    Focus relationship perception for unsupervised multi-focus image fusion,

    J. Liu, S. Li, R. Dian, and Z. Song, “Focus relationship perception for unsupervised multi-focus image fusion,”IEEE Transactions on Multimedia, vol. 26, pp. 6155–6165, 2023

    2023

  34. [42]

    Model-guided coarse- to-fine fusion network for unsupervised hyperspectral image super- resolution,

    J. Li, K. Zheng, W. Liu, Z. Li, H. Yu, and L. Ni, “Model-guided coarse- to-fine fusion network for unsupervised hyperspectral image super- resolution,”IEEE Geoscience and Remote Sensing Letters, vol. 20, pp. 1–5, 2023

    2023

  35. [43]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595

    2018

  36. [44]

    Crossfuse: A novel cross attention mechanism based infrared and visible image fusion approach,

    H. Li and X.-J. Wu, “Crossfuse: A novel cross attention mechanism based infrared and visible image fusion approach,”Information Fusion, vol. 103, p. 102147, 2024

    2024

  37. [45]

    Enhanced deep image prior for unsupervised hyperspectral image super-resolution,

    J. Li, K. Zheng, L. Gao, Z. Han, Z. Li, and J. Chanussot, “Enhanced deep image prior for unsupervised hyperspectral image super-resolution,” IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–18, 2025

    2025

  38. [46]

    Multiscale vision transformers,

    H. Fan, B. Xiong, K. Mangalam, Y . Li, Z. Yan, J. Malik, and C. Feichten- hofer, “Multiscale vision transformers,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 6824–6835

    2021

  39. [47]

    Nafssr: Stereo image super-resolution using nafnet,

    X. Chu, L. Chen, and W. Yu, “Nafssr: Stereo image super-resolution using nafnet,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 1239–1248

    2022

  40. [48]

    Anα-matte boundary defocus model-based cascaded network for multi-focus image fusion,

    H. Ma, Q. Liao, J. Zhang, S. Liu, and J.-H. Xue, “Anα-matte boundary defocus model-based cascaded network for multi-focus image fusion,” IEEE Transactions on Image Processing, vol. 29, pp. 8668–8679, 2020

    2020

  41. [49]

    Rgb-d saliency detection via cascaded mutual information minimiza- tion,

    J. Zhang, D.-P. Fan, Y . Dai, X. Yu, Y . Zhong, N. Barnes, and L. Shao, “Rgb-d saliency detection via cascaded mutual information minimiza- tion,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 4338–4347

    2021

  42. [50]

    Multi-focus image fusion using dictionary-based sparse representation,

    M. Nejati, S. Samavi, and S. Shirani, “Multi-focus image fusion using dictionary-based sparse representation,”Information fusion, vol. 25, pp. 72–84, 2015

    2015

  43. [51]

    Mffw: A new dataset for multi-focus image fusion,

    S. Xu, X. Wei, C. Zhang, J. Liu, and J. Zhang, “Mffw: A new dataset for multi-focus image fusion,”arXiv preprint arXiv:2002.04780, 2020

    2002 arXiv

  44. [52]

    Mff-gan: An unsuper- vised generative adversarial network with adaptive and gradient joint constraints for multi-focus image fusion,

    H. Zhang, Z. Le, Z. Shao, H. Xu, and J. Ma, “Mff-gan: An unsuper- vised generative adversarial network with adaptive and gradient joint constraints for multi-focus image fusion,”Information Fusion, vol. 66, pp. 40–53, 2021

    2021

  45. [53]

    Standard images for multifocus image fusion,

    C.-C. Tsai, “Standard images for multifocus image fusion,” 2014

    2014

  46. [54]

    Unfolding coupled convolutional sparse representation for multi-focus image fusion,

    K. Zheng, J. Cheng, and Y . Liu, “Unfolding coupled convolutional sparse representation for multi-focus image fusion,”Information Fusion, vol. 118, p. 102974, 2025

    2025

  47. [55]

    Multi-focus image fusion with dense sift,

    Y . Liu, S. Liu, and Z. Wang, “Multi-focus image fusion with dense sift,” Information Fusion, vol. 23, pp. 139–155, 2015

    2015

  48. [56]

    Multi-focus image fusion based on multi-scale focus measures and generalized random walk,

    J. Ma, Z. Zhou, B. Wang, and M. Dong, “Multi-focus image fusion based on multi-scale focus measures and generalized random walk,” in2017 36th Chinese control conference (CCC). IEEE, 2017, pp. 5464–5468

    2017

  49. [57]

    Multi-focus image fusion using singular value decomposition in dct domain,

    M. Amin-Naji, P. Ranjbar-Noiey, and A. Aghagolzadeh, “Multi-focus image fusion using singular value decomposition in dct domain,” in 2017 10th Iranian conference on machine vision and image processing (MVIP). IEEE, 2017, pp. 45–51

    2017

  50. [58]

    Sesf- fuse: An unsupervised deep model for multi-focus image fusion,

    B. Ma, Y . Zhu, X. Yin, X. Ban, H. Huang, and M. Mukeshimana, “Sesf- fuse: An unsupervised deep model for multi-focus image fusion,”Neural Computing and Applications, vol. 33, no. 11, pp. 5793–5804, 2021

    2021

  51. [59]

    End-to-end learning for simultaneously generating decision map and multi-focus image fusion result,

    B. Ma, X. Yin, D. Wu, H. Shen, X. Ban, and Y . Wang, “End-to-end learning for simultaneously generating decision map and multi-focus image fusion result,”Neurocomputing, vol. 470, pp. 204–216, 2022

    2022

  52. [60]

    Mufusion: A general unsupervised image fusion network based on memory unit,

    C. Cheng, T. Xu, and X.-J. Wu, “Mufusion: A general unsupervised image fusion network based on memory unit,”Information Fusion, vol. 92, pp. 80–92, 2023

    2023

  53. [61]

    Fusiondiff: Multi-focus image fusion using denoising diffusion probabilistic models,

    M. Li, R. Pei, T. Zheng, Y . Zhang, and W. Fu, “Fusiondiff: Multi-focus image fusion using denoising diffusion probabilistic models,”Expert Systems with Applications, vol. 238, p. 121664, 2024

    2024

  54. [62]

    Multi-focus image fusion via interactive transformer and asymmetric soft sharing,

    H. Zhai, W. Zheng, Y . Ouyang, X. Pan, and W. Zhang, “Multi-focus image fusion via interactive transformer and asymmetric soft sharing,” Engineering Applications of Artificial Intelligence, vol. 133, p. 107967, 2024

    2024

  55. [63]

    Sd- fuse: An image structure-driven model for multi-focus image fusion,

    Z. Wang, J. Wang, H. Song, P. Wang, K. Lyu, W. Li, and L. Zhao, “Sd- fuse: An image structure-driven model for multi-focus image fusion,” Information Fusion, p. 104058, 2025

    2025

  56. [64]

    Comments on ‘infor- mation measure for performance of image fusion’,

    M. Hossny, S. Nahavandi, and D. Creighton, “Comments on ‘infor- mation measure for performance of image fusion’,”Electronics letters, vol. 44, no. 18, pp. 1066–1067, 2008

    2008

  57. [65]

    A nonlinear correlation measure for multivariable data set,

    Q. Wang, Y . Shen, and J. Q. Zhang, “A nonlinear correlation measure for multivariable data set,”Physica D: Nonlinear Phenomena, vol. 200, no. 3-4, pp. 287–295, 2005

    2005

  58. [66]

    Objective image fusion performance measure,

    C. S. Xydeas and V . Petrovic, “Objective image fusion performance measure,”Electronics letters, vol. 36, no. 4, pp. 308–309, 2000

    2000

  59. [67]

    A novel similarity based quality metric for image fusion,

    C. Yang, J.-Q. Zhang, X.-R. Wang, and X. Liu, “A novel similarity based quality metric for image fusion,”Information Fusion, vol. 9, no. 2, pp. 156–160, 2008

    2008

  60. [68]

    A similarity metric for assessment of image fusion algorithms,

    N. Cvejic, A. Loza, D. Bull, and N. Canagarajah, “A similarity metric for assessment of image fusion algorithms,”International journal of signal processing, vol. 2, no. 3, pp. 178–182, 2005

    2005

  61. [69]

    A new automated quality assessment algorithm for image fusion,

    Y . Chen and R. S. Blum, “A new automated quality assessment algorithm for image fusion,”Image and vision computing, vol. 27, no. 10, pp. 1421–1432, 2009

    2009

  62. [70]

    Rethinking the effectiveness of objective evaluation metrics in multi-focus image fusion: A statistic- based approach,

    Y . Liu, Z. Qi, J. Cheng, and X. Chen, “Rethinking the effectiveness of objective evaluation metrics in multi-focus image fusion: A statistic- based approach,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 8, pp. 5806–5819, 2024

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.