pith. sign in

arxiv: 2509.11485 · v2 · submitted 2025-09-15 · ❄️ cond-mat.mtrl-sci · cs.CV

Geometric Analysis of Magnetic Labyrinthine Stripe Evolution via U-Net Segmentation

Vin\'icius Yu Okubo , Kotaro Shimizu , B.S. Shivaran , Gia-Wei Chern , Hae Yong Kim This is my paper

Pith reviewed 2026-05-18 17:11 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci cs.CV
keywords magnetic labyrinthine stripesU-Net segmentationBi:YIG filmsmagnetic annealinggeometric analysisstripe evolution modesmagneto-optical imaging
0
0 comments X

The pith

U-Net segmentation of magneto-optical images enables geometric tracking of magnetic stripe evolution and reveals two polarity-linked modes during annealing.

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

The paper aims to quantitatively characterize how labyrinthine magnetic stripe patterns in Bi:YIG films evolve from a disordered quenched state to a more ordered annealed state under magnetic field annealing. A U-Net model is trained on synthetic noise to segment noisy real images, followed by skeletonization and spline fitting to extract local stripe lengths and curvatures. Applied to 444 images across 12 trials, this pipeline identifies two distinct evolution modes, Type A and Type B, that depend on the polarity of the applied field. A sympathetic reader would care because the lack of long-range order in such patterns has made precise geometric and topological measurements difficult, and the method offers a scalable way to study their local dynamics.

Core claim

A U-Net trained exclusively on synthetic degradations including additive white Gaussian and Simplex noise segments experimental magneto-optical images of magnetic stripes in Bi:YIG films with sufficient accuracy for downstream analysis. Skeletonization, graph mapping, and spline fitting then quantify local stripe propagation through length and curvature. When applied to 444 images from 12 annealing protocol trials, the measurements show a transition from the quenched state to a more parallel annealed state and distinguish two evolution modes, Type A and Type B, that are linked to field polarity.

What carries the argument

The U-Net model for segmenting noisy magneto-optical images, combined with skeletonization, graph mapping, and spline fitting to extract stripe length and curvature measurements.

If this is right

  • The method quantifies the geometric and topological changes during the transition from quenched to annealed states.
  • Two distinct evolution modes, Type A and Type B, are linked to the polarity of the applied magnetic field.
  • Local measurements of stripe length and curvature provide concrete data on structural evolution in labyrinthine patterns.
  • The pipeline establishes a general tool for analyzing complex labyrinthine systems beyond this specific material.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same segmentation-plus-geometry approach could be applied to time-series data to track individual stripe motion rather than ensemble statistics.
  • If the two modes prove robust, targeted field protocols might be designed to favor one mode for applications requiring specific pattern alignments.
  • Similar deep-learning segmentation could be tested on other imaging modalities in materials science where noise and occlusions obscure linear features.

Load-bearing premise

The U-Net model trained exclusively on synthetic degradations produces segmentations of real experimental images that are sufficiently accurate and unbiased for reliable geometric measurements of length and curvature.

What would settle it

Manually annotating a representative subset of the 444 experimental images, recomputing the length and curvature statistics, and checking whether the reported differences between Type A and Type B modes or between quenched and annealed states remain statistically significant.

Figures

Figures reproduced from arXiv: 2509.11485 by B.S. Shivaran, Gia-Wei Chern, Hae Yong Kim, Kotaro Shimizu, Vin\'icius Yu Okubo.

Figure 1
Figure 1. Figure 1: FIGURE 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIGURE 2 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIGURE 3 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIGURE 4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: shows the outcome: yellow marks the low-noise regions used for training, while purple highlights the high￾noise areas that were excluded. In order to make the model capable at segmenting the de￾graded regions of the magnetic stripe pattern, we synthetically introduced degradations during training to the labyrinthine patches using Gaussian and Simplex noises [59]. Simplex noise allowed for the creation of s… view at source ↗
Figure 7
Figure 7. Figure 7: illustrates the process: (a) shows the original mag￾netic stripe pattern with TM-CNN-detected defects high￾lighted; (b) shows the untrimmed skeleton; and (c) shows the result after trimming. The boundary of the stripe patterns were produced through contour finding using OpenCV’s findContours, which is based on the Suzuki-Abe algorithm [60] [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIGURE 8 [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIGURE 9 [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Each mean-length measurement exhibits a distinct [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIGURE 10 [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 13
Figure 13. Figure 13: (a)-(b) illustrates the total inner and border lengths of the magnetic stripe pattern images. The graphs show a lower average length at the beginning of the annealing protocol (in the quenched state), followed by stabilization at a higher average length in the later steps (in the annealed state). This increase in total length throughout the annealing protocol can be attributed to a more efficient packagin… view at source ↗
Figure 14
Figure 14. Figure 14: FIGURE 14 [PITH_FULL_IMAGE:figures/full_fig_p015_14.png] view at source ↗
read the original abstract

Labyrinthine stripe patterns are common in many physical systems, yet their lack of long-range order makes quantitative characterization challenging. We investigate the evolution of such patterns in bismuth-doped yttrium iron garnet (Bi:YIG) films subjected to a magnetic field annealing protocol. A U-Net deep learning model, trained with synthetic degradations including additive white Gaussian and Simplex noise, enables robust segmentation of experimental magneto-optical images despite noise and occlusions. Building on this segmentation, we develop a geometric analysis pipeline based on skeletonization, graph mapping, and spline fitting, which quantifies local stripe propagation through length and curvature measurements. Applying this framework to 444 images from 12 annealing protocol trials, we analyze the transition from the "quenched" state to a more parallel and coherent "annealed" state, and identify two distinct evolution modes (Type A and Type B) linked to field polarity. Our results provide a quantitative analysis of geometric and topological properties in magnetic stripe patterns and offer new insights into their local structural evolution, and establish a general tool for analyzing complex labyrinthine systems.

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.

Referee Report

2 major / 2 minor

Summary. The paper develops a U-Net segmentation model trained exclusively on synthetic additive white Gaussian and Simplex noise degradations to process magneto-optical images of labyrinthine stripe patterns in Bi:YIG films. It then applies a geometric pipeline (skeletonization, graph mapping, spline fitting) to extract length and curvature metrics from 444 images across 12 annealing trials, identifying a transition from quenched to annealed states and two distinct evolution modes (Type A and Type B) linked to field polarity.

Significance. If the segmentation step proves accurate and unbiased on real data, the work supplies a practical quantitative toolkit for characterizing disordered magnetic stripe patterns, enabling reproducible measurements of local propagation and topology that are otherwise difficult in labyrinthine systems. The scale of the experimental dataset (444 images) and the identification of polarity-linked modes represent a concrete advance in applying image-based geometric analysis to materials-science problems.

major comments (2)
  1. [Methods (U-Net subsection)] Methods section on U-Net training and validation: The model is trained solely on synthetic noise degradations, yet the manuscript reports no quantitative metrics (IoU, Dice, boundary F1, or pixel-wise error) and no direct comparison against manual annotations on actual Bi:YIG magneto-optical images. Because the central claims—identification of Type A versus Type B modes and the quenched-to-annealed transition—rest entirely on downstream length and curvature statistics derived from these segmentations, the absence of real-image validation is load-bearing and must be addressed before the geometric conclusions can be trusted.
  2. [Results (mode identification)] Results section describing mode classification: The distinction between Type A and Type B evolution modes is asserted to be linked to field polarity, but without reported error bars, statistical tests, or sensitivity analysis on the curvature/length distributions, it is unclear whether the separation survives plausible segmentation errors (e.g., boundary shifts of a few pixels). A concrete test—re-running the pipeline on a small manually annotated subset and showing that mode labels remain stable—would be required to substantiate the claim.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'robust segmentation … despite noise and occlusions' is stated without supporting numbers; a brief parenthetical reference to any internal synthetic-test metrics would improve clarity.
  2. [Throughout] Figure captions and text: Several instances of undefined abbreviations (e.g., exact definition of 'quenched' versus 'annealed' state thresholds) appear; a short glossary or explicit definition in the first use would aid readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments on our manuscript. We have addressed each major comment point by point below, with revisions incorporated where appropriate to strengthen the validation and robustness of our findings.

read point-by-point responses

  1. Referee: [Methods (U-Net subsection)] Methods section on U-Net training and validation: The model is trained solely on synthetic noise degradations, yet the manuscript reports no quantitative metrics (IoU, Dice, boundary F1, or pixel-wise error) and no direct comparison against manual annotations on actual Bi:YIG magneto-optical images. Because the central claims—identification of Type A versus Type B modes and the quenched-to-annealed transition—rest entirely on downstream length and curvature statistics derived from these segmentations, the absence of real-image validation is load-bearing and must be addressed before the geometric conclusions can be trusted.

    Authors: We acknowledge that quantitative validation on real experimental images is important to support the reliability of the segmentation step and the downstream geometric conclusions. The original manuscript presented qualitative visual comparisons of segmented outputs on experimental magneto-optical images to demonstrate performance. To directly address this point, the revised manuscript now includes a dedicated validation subsection in Methods. This reports quantitative metrics (IoU and Dice scores) computed against manual annotations on a subset of real Bi:YIG images, along with a direct comparison to establish accuracy. These additions confirm that the segmentation quality is sufficient for the reported length and curvature statistics. revision: yes

  2. Referee: [Results (mode identification)] Results section describing mode classification: The distinction between Type A and Type B evolution modes is asserted to be linked to field polarity, but without reported error bars, statistical tests, or sensitivity analysis on the curvature/length distributions, it is unclear whether the separation survives plausible segmentation errors (e.g., boundary shifts of a few pixels). A concrete test—re-running the pipeline on a small manually annotated subset and showing that mode labels remain stable—would be required to substantiate the claim.

    Authors: We agree that additional statistical rigor and sensitivity checks are needed to substantiate the mode distinction and its link to field polarity. In the revised manuscript, we have added error bars to the length and curvature distribution plots (representing variation across trials) and included statistical tests (two-sample t-tests) to quantify the significance of differences between Type A and Type B modes. We have also performed a sensitivity analysis by applying controlled boundary perturbations to the segmentations and re-running the pipeline, confirming that mode assignments remain stable. Furthermore, we re-ran the full geometric analysis on a small manually annotated subset of real images and verified that the identified modes are consistent with those from the larger dataset. These results are now reported in the Results section with an accompanying supplementary figure. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical pipeline is self-contained data analysis

full rationale

The paper describes an applied image-processing workflow: a U-Net is trained exclusively on synthetic degradations and then used to segment real experimental magneto-optical images, after which skeletonization, graph mapping, and spline fitting produce length and curvature statistics that are inspected across 444 frames to label Type A versus Type B evolution modes. No equation or claim reduces a reported result to the same fitted quantity by construction, no uniqueness theorem is imported via self-citation, and no ansatz is smuggled in. The central observations are direct measurements on the experimental dataset and therefore remain independent of the inputs used to train the segmenter.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the generalization of a U-Net trained only on synthetic noise to real experimental images and on the assumption that skeletonization faithfully captures the topological and geometric properties of the physical stripes without introducing systematic bias.

axioms (2)
  • domain assumption Synthetic noise degradations (additive white Gaussian and Simplex) are sufficient to train a model that generalizes to real magneto-optical images.
    Stated in the abstract as the training strategy; no real-image ground truth is mentioned.
  • domain assumption Skeletonization followed by graph mapping and spline fitting produces length and curvature values that accurately reflect physical stripe propagation.
    Implicit in the geometric analysis pipeline description.

pith-pipeline@v0.9.0 · 5736 in / 1545 out tokens · 48174 ms · 2026-05-18T17:11:49.186154+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages · 1 internal anchor

  1. [1]

    Nicolis and I

    G. Nicolis and I. Prigogine,Self Organization in Non Equilibrium Systems. Wiley, 1977

    work page 1977

  2. [2]

    M. C. Cross and P . C. Hohenberg, ‘‘Pattern formation outside of equilibrium,’’Rev. Mod. Phys., vol. 65, pp. 851–1112, Jul 1993. [Online]. Available: https://link.aps.org/doi/10.1103/RevModPhys.65.851

    work page doi:10.1103/revmodphys.65.851 1993

  3. [3]

    A. J. Koch and H. Meinhardt, ‘‘Biological pattern formation: from basic mechanisms to complex structures,’’Rev. Mod. Phys., vol. 66, pp. 1481–1507, Oct 1994. [Online]. Available: https://link.aps.org/doi/10. 1103/RevModPhys.66.1481

    work page 1994

  4. [4]

    Pismen,Patterns and Interfaces in Dissipative Dynamics

    L. Pismen,Patterns and Interfaces in Dissipative Dynamics. Springer, 2006

    work page 2006

  5. [5]

    Cross and H

    M. Cross and H. Greenside,Pattern F ormation and Dynamics in Non- Equilibrium Systems. Cambridge University Press, 2009

    work page 2009

  6. [6]

    Ouyang and H

    Q. Ouyang and H. L. Swinney, ‘‘Transition from a uniform state to hexagonal and striped turing patterns,’’Nature, vol. 352, no. 6336, pp. 610– 612, Aug 1991. [Online]. Available: https://doi.org/10.1038/352610a0

    work page doi:10.1038/352610a0 1991

  7. [7]

    K. R. Elder, J. Viñals, and M. Grant, ‘‘Dynamic scaling and quasiordered states in the two-dimensional swift-hohenberg equation,’’Phys. Rev. A, vol. 46, pp. 7618–7629, Dec 1992. [Online]. Available: https: //link.aps.org/doi/10.1103/PhysRevA.46.7618

    work page doi:10.1103/physreva.46.7618 1992

  8. [8]

    M. C. Cross and D. I. Meiron, ‘‘Domain coarsening in systems far from equilibrium,’’Phys. Rev. Lett., vol. 75, pp. 2152–2155, Sep 1995. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevLett.75.2152

    work page doi:10.1103/physrevlett.75.2152 1995

  9. [9]

    J. J. Christensen and A. J. Bray, ‘‘Pattern dynamics of rayleigh- bénard convective rolls and weakly segregated diblock copolymers,’’ Phys. Rev. E, vol. 58, pp. 5364–5370, Nov 1998. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevE.58.5364

    work page doi:10.1103/physreve.58.5364 1998

  10. [10]

    Echeverría-Alar and M

    S. Echeverría-Alar and M. G. Clerc, ‘‘Labyrinthine patterns transitions,’’ Phys. Rev. Res., vol. 2, p. 042036, Dec 2020. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevResearch.2.042036

    work page doi:10.1103/physrevresearch.2.042036 2020

  11. [11]

    G. H. Gunaratne, R. E. Jones, Q. Ouyang, and H. L. Swinney, ‘‘An invariant measure of disorder in patterns,’’Phys. Rev. Lett., vol. 75, pp. 3281–3284, Oct 1995. [Online]. Available: https://link.aps.org/doi/10. 1103/PhysRevLett.75.3281

    work page 1995

  12. [12]

    S. Hu, D. I. Goldman, D. J. Kouri, D. K. Hoffman, H. L. Swinney, and G. H. Gunaratne, ‘‘Stages of relaxation of patterns and the role of stochasticity in the final stage,’’Nonlinearity, vol. 17, no. 4, p. 1535, may

    work page

  13. [13]

    Available: https://dx.doi.org/10.1088/0951-7715/17/4/021

    [Online]. Available: https://dx.doi.org/10.1088/0951-7715/17/4/021

    work page doi:10.1088/0951-7715/17/4/021

  14. [14]

    Nathan and G

    G. Nathan and G. Gunaratne, ‘‘Set of measures to analyze the dynamics of nonequilibrium structures,’’Phys. Rev. E, vol. 71, p. 035101, Mar 2005. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevE.71.035101

    work page doi:10.1103/physreve.71.035101 2005

  15. [15]

    S. Hu, G. Nathan, D. J. Kouri, D. K. Hoffman, and G. H. Gunaratne, ‘‘Statistical characterizations of spatiotemporal patterns generated in the Swift–Hohenberg model,’’Chaos: An Interdisciplinary Journal of Nonlinear Science, vol. 15, no. 4, p. 043701, 11 2005. [Online]. Available: https://doi.org/10.1063/1.2046487

    work page doi:10.1063/1.2046487 2005

  16. [16]

    Q. Hou, S. Sasa, and N. Goldenfeld, ‘‘Dynamical scaling behavior of the swift-hohenberg equation following a quench to the modulated state,’’ Physica A: Statistical Mechanics and its Applications, vol. 239, no. 1, pp. 219–226, 1997. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S0378437196004803

    work page 1997

  17. [17]

    Seul and R

    M. Seul and R. Wolfe, ‘‘Evolution of disorder in two-dimensional stripe patterns: ‘‘smectic’’ instabilities and disclination unbinding,’’Phys. Rev. Lett., vol. 68, pp. 2460–2463, Apr 1992. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevLett.68.2460

    work page doi:10.1103/physrevlett.68.2460 1992

  18. [18]

    ——, ‘‘Evolution of disorder in magnetic stripe domains. ii. hairpins and labyrinth patterns versus branches and comb patterns formed by growing minority component,’’Phys. Rev. A, vol. 46, pp. 7534–7547, Dec 1992. [Online]. Available: https://link.aps.org/doi/10.1103/PhysRevA.46.7534

    work page doi:10.1103/physreva.46.7534 1992

  19. [19]

    L. R. M. M. Seul and L. O’Gorman, ‘‘Pattern analysis of magnetic stripe domains morphology and topological defects in the disordered state,’’ Philosophical Magazine B, vol. 66, no. 4, pp. 471–506, 1992. [Online]. Available: https://doi.org/10.1080/13642819208220116

    work page doi:10.1080/13642819208220116 1992

  20. [20]

    V . Y . Okubo, K. Shimizu, B. S. Shivaram, and H. Y ong Kim, ‘‘Characteri- zation of magnetic labyrinthine structures through junctions and terminals detection using template matching and cnn,’’IEEE Access, vol. 12, pp. 92 419–92 430, 2024

    work page 2024

  21. [21]

    Shimizu, V

    K. Shimizu, V . Y . Okubo, R. Knight, Z. Wang, J. Burton, H. Y . Kim, G.-W. Chern, and B. Shivaram, ‘‘Machine learning assisted characterization of labyrinthine pattern transitions,’’arXiv preprint arXiv:2311.10558, 2023

    work page arXiv 2023

  22. [22]

    J. Long, E. Shelhamer, and T. Darrell, ‘‘Fully convolutional networks for semantic segmentation,’’ inProceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 3431–3440

    work page 2015

  23. [23]

    Ronneberger, P

    O. Ronneberger, P . Fischer, and T. Brox, ‘‘U-net: Convolutional networks for biomedical image segmentation,’’ inMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international confer- ence, Munich, Germany, October 5-9, 2015, proceedings, part III 18. Springer, 2015, pp. 234–241

    work page 2015

  24. [24]

    R. Azad, E. K. Aghdam, A. Rauland, Y . Jia, A. H. Avval, A. Bozorgpour, S. Karimijafarbigloo, J. P . Cohen, E. Adeli, and D. Merhof, ‘‘Medical image segmentation review: The success of u-net,’’IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    work page 2024

  25. [25]

    J. Wu, W. Liu, C. Li, T. Jiang, I. M. Shariful, Y . Y ao, H. Sun, X. Li, X. Li, X. Huang, and M. Grzegorzek, ‘‘A state-of-the-art survey of U-Net in microscopic image analysis: from simple usage to structure mortification,’’ Neural Computing and Applications, vol. 36, no. 7, pp. 3317–3346, Mar. 2024

    work page 2024

  26. [26]

    W. Xia, Z. Cheng, Y . Y ang, and J.-H. Xue, ‘‘Cooperative semantic segmen- tation and image restoration in adverse environmental conditions,’’arXiv preprint arXiv:1911.00679, 2019

    work page arXiv 1911

  27. [27]

    D. Guo, Y . Pei, K. Zheng, H. Y u, Y . Lu, and S. Wang, ‘‘Degraded image semantic segmentation with dense-gram networks,’’IEEE Transactions on Image Processing, vol. 29, pp. 782–795, 2019

    work page 2019

  28. [28]

    A. Rajagopalanet al., ‘‘Improving robustness of semantic segmentation to motion-blur using class-centric augmentation,’’ inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 10 470–10 479

    work page 2023

  29. [29]

    F. Zhao, J. Feng, J. Zhao, W. Y ang, and S. Y an, ‘‘Robust lstm-autoencoders for face de-occlusion in the wild,’’IEEE Transactions on Image Process- ing, vol. 27, no. 2, pp. 778–790, 2017

    work page 2017

  30. [30]

    Matsui and M

    T. Matsui and M. Ikehara, ‘‘Single-image fence removal using deep con- volutional neural network,’’IEEE Access, vol. 8, pp. 38 846–38 854, 2019

    work page 2019

  31. [31]

    Sakaridis, D

    C. Sakaridis, D. Dai, and L. V an Gool, ‘‘Semantic foggy scene understand- ing with synthetic data,’’International Journal of Computer Vision, vol. 126, pp. 973–992, 2018

    work page 2018

  32. [32]

    Z. Hao, S. Y ou, Y . Li, K. Li, and F. Lu, ‘‘Learning from synthetic photo- realistic raindrop for single image raindrop removal,’’ inProceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 2019, pp. 0–0

    work page 2019

  33. [33]

    H. C. Burger, C. J. Schuler, and S. Harmeling, ‘‘Image denoising: Can plain neural networks compete with bm3d?’’ in2012 IEEE conference on computer vision and pattern recognition. IEEE, 2012, pp. 2392–2399

    work page 2012

  34. [34]

    Zhang, W

    K. Zhang, W. Zuo, Y . Chen, D. Meng, and L. Zhang, ‘‘Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising,’’IEEE trans- actions on image processing, vol. 26, no. 7, pp. 3142–3155, 2017

    work page 2017

  35. [35]

    Zhang, W

    K. Zhang, W. Zuo, and L. Zhang, ‘‘Ffdnet: Toward a fast and flexible solution for cnn-based image denoising,’’IEEE Transactions on Image Processing, vol. 27, no. 9, pp. 4608–4622, 2018

    work page 2018

  36. [36]

    B. Park, S. Y u, and J. Jeong, ‘‘Densely connected hierarchical network for image denoising,’’ inProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2019, pp. 0–0

    work page 2019

  37. [37]

    Bae, C.-W

    H.-J. Bae, C.-W. Kim, N. Kim, B. Park, N. Kim, J. B. Seo, and S. M. Lee, ‘‘A perlin noise-based augmentation strategy for deep learning with small 12 VOLUME 11, 2023 Okuboet al.: Geometric Analysis of Magnetic Labyrinthine Stripe Evolution data samples of hrct images,’’Scientific Reports, vol. 8, no. 1, p. 17687, Dec

    work page 2023

  38. [38]

    Available: https://doi.org/10.1038/s41598-018-36047-2

    [Online]. Available: https://doi.org/10.1038/s41598-018-36047-2

    work page doi:10.1038/s41598-018-36047-2

  39. [39]

    Haekal, R

    M. Haekal, R. Septiawan, F. Haryanto, and I. Arif, ‘‘A comparison on the use of perlin-noise and gaussian noise based augmentation on x-ray classification of lung cancer patient,’’Journal of Physics: Conference Series, vol. 1951, no. 1, p. 012064, jun 2021. [Online]. Available: https://dx.doi.org/10.1088/1742-6596/1951/1/012064

    work page doi:10.1088/1742-6596/1951/1/012064 1951

  40. [40]

    Zheng, J

    Y . Zheng, J. Su, S. Zhang, M. Tao, and L. Wang, ‘‘Unpaired remote sensing image dehazing using enhanced skip attention-based generative adversarial networks with rotation invariance,’’Remote Sensing, vol. 16, no. 15, 2024. [Online]. Available: https://www.mdpi.com/2072-4292/16/15/2707

    work page 2024

  41. [41]

    Mahmood, M

    F. Mahmood, M. Inglezou, P . Chatzakos, and A. Porichis, ‘‘Enhancing crack segmentation with limited data: Singan-based synthesis and blending of textures and cracks,’’IEEE Access, vol. 13, pp. 104 555–104 564, 2025

    work page 2025

  42. [42]

    J. Li, M. Li, S. Huang, G. Wang, and X. Zhao, ‘‘Industrial image anomaly detection via synthetic-anomaly contrastive distillation,’’Sensors, vol. 25, no. 12, 2025. [Online]. Available: https://www.mdpi.com/1424-8220/25/ 12/3721

    work page 2025

  43. [43]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy, ‘‘An image is worth 16x16 words: Transformers for image recognition at scale,’’arXiv preprint arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  44. [44]

    Strudel, R

    R. Strudel, R. Garcia, I. Laptev, and C. Schmid, ‘‘Segmenter: Transformer for semantic segmentation,’’ inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 7262–7272

    work page 2021

  45. [45]

    Zheng, J

    S. Zheng, J. Lu, H. Zhao, X. Zhu, Z. Luo, Y . Wang, Y . Fu, J. Feng, T. Xiang, P . H. Torret al., ‘‘Rethinking semantic segmentation from a sequence-to- sequence perspective with transformers,’’ inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 6881– 6890

    work page 2021

  46. [46]

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, ‘‘Swin transformer: Hierarchical vision transformer using shifted windows,’’ in Proceedings of the IEEE/CVF international conference on computer vi- sion, 2021, pp. 10 012–10 022

    work page 2021

  47. [47]

    H. Chen, Y . Wang, T. Guo, C. Xu, Y . Deng, Z. Liu, S. Ma, C. Xu, C. Xu, and W. Gao, ‘‘Pre-trained image processing transformer,’’ inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 12 299–12 310

    work page 2021

  48. [48]

    Liang, J

    J. Liang, J. Cao, G. Sun, K. Zhang, L. V an Gool, and R. Timofte, ‘‘Swinir: Image restoration using swin transformer,’’ inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 1833– 1844

    work page 2021

  49. [49]

    S. W. Zamir, A. Arora, S. Khan, M. Hayat, F. S. Khan, and M.-H. Y ang, ‘‘Restormer: Efficient transformer for high-resolution image restoration,’’ inProceedings of the IEEE/CVF conference on computer vision and pat- tern recognition, 2022, pp. 5728–5739

    work page 2022

  50. [50]

    Isola, J.-Y

    P . Isola, J.-Y . Zhu, T. Zhou, and A. A. Efros, ‘‘Image-to-image translation with conditional adversarial networks,’’ inProceedings of the IEEE con- ference on computer vision and pattern recognition, 2017, pp. 1125–1134

    work page 2017

  51. [51]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde- Farley, S. Ozair, A. Courville, and Y . Bengio, ‘‘Generative adversarial nets,’’ inAdvances in Neural Information Processing Systems, Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, Eds., vol. 27. Curran Associates, Inc.,

    work page

  52. [52]

    Available: https://proceedings.neurips.cc/paper_files/ paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf

    [Online]. Available: https://proceedings.neurips.cc/paper_files/ paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf

    work page 2014

  53. [53]

    Majurski, P

    M. Majurski, P . Manescu, S. Padi, N. Schaub, N. Hotaling, C. Simon Jr, and P . Bajcsy, ‘‘Cell image segmentation using generative adversarial networks, transfer learning, and augmentations,’’ inProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2019, pp. 0–0

    work page 2019

  54. [54]

    Kawar, M

    B. Kawar, M. Elad, S. Ermon, and J. Song, ‘‘Denoising diffusion restora- tion models,’’Advances in neural information processing systems, vol. 35, pp. 23 593–23 606, 2022

    work page 2022

  55. [55]

    Z. Luo, F. K. Gustafsson, Z. Zhao, J. Sjölund, and T. B. Schön, ‘‘Image restoration with mean-reverting stochastic differential equations,’’arXiv preprint arXiv:2301.11699, 2023

    work page arXiv 2023

  56. [56]

    Y . Xie, M. Y uan, B. Dong, and Q. Li, ‘‘Diffusion model for generative image denoising,’’arXiv preprint arXiv:2302.02398, 2023

    work page arXiv 2023

  57. [57]

    R. Qian, R. T. Tan, W. Y ang, J. Su, and J. Liu, ‘‘Attentive generative adver- sarial network for raindrop removal from a single image,’’ inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 2482–2491

    work page 2018

  58. [58]

    J. Dong, L. Zhang, H. Zhang, and W. Liu, ‘‘Occlusion-aware gan for face de-occlusion in the wild,’’ in2020 IEEE International conference on multimedia and expo (ICME). IEEE Computer Society, 2020, pp. 1–6

    work page 2020

  59. [59]

    Angah and A

    O. Angah and A. Y . Chen, ‘‘Removal of occluding construction workers in job site image data using u-net based context encoders,’’Automation in Construction, vol. 119, p. 103332, 2020

    work page 2020

  60. [60]

    I. B. Puchalska, G. A. Jones, and H. Jouve, ‘‘A new aspect on the observation of domain structure in garnet epilayers,’’Journal of Physics D: Applied Physics, vol. 11, no. 15, p. L175, oct 1978. [Online]. Available: https://dx.doi.org/10.1088/0022-3727/11/15/002

    work page doi:10.1088/0022-3727/11/15/002 1978

  61. [61]

    Otsu, ‘‘A threshold selection method from gray-level histograms,’’IEEE Transactions on Systems, Man, and Cybernetics, vol

    N. Otsu, ‘‘A threshold selection method from gray-level histograms,’’IEEE Transactions on Systems, Man, and Cybernetics, vol. 9, no. 1, pp. 62–66, 1979

    work page 1979

  62. [62]

    Perlin, ‘‘Improving noise,’’ inProceedings of the 29th annual confer- ence on Computer graphics and interactive techniques, 2002, pp

    K. Perlin, ‘‘Improving noise,’’ inProceedings of the 29th annual confer- ence on Computer graphics and interactive techniques, 2002, pp. 681–682

    work page 2002

  63. [63]

    Suzuki and K

    S. Suzuki and K. Abe, ‘‘Topological structural analysis of digitized binary images by border following,’’Computer Vision, Graphics, and Image Processing, vol. 30, no. 1, pp. 32–46, 1985. [Online]. Available: https://www.sciencedirect.com/science/article/pii/0734189X85900167

    work page arXiv 1985

  64. [64]

    html#smoothing-splines-with-automatic-knot-selection, accessed: 2025- 08-24

    SciPy Community, ‘‘Scipy v1.16.1 reference guide: Smoothing splines,’’ https://docs.scipy.org/doc/scipy/tutorial/interpolate/smoothing_splines. html#smoothing-splines-with-automatic-knot-selection, accessed: 2025- 08-24. VINÍCIUS YU OKUBOwas born in Campinas, Brazil, in 1999. He completed his B.S. in electri- cal engineering from the University of São Pau...

    work page 2025