Pith. sign in

REVIEW 2 major objections 5 minor 53 references

Rotation-Parameterized Graph Fractional Fourier Transform: Definition, Properties, and Optimal Filtering

T0 review · 2 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

Pith's one-line read A new graph spectral transform unifies fractional order and rotation angle by fixing the zero-angle degeneracy flaw of angular graph Fourier transforms.

desk verdict The transform construction is coherent and the zero-angle fix is real, but the denoising results are training losses, not predictions. read the letter →

arxiv 2511.16111 v2 pith:RYZI4TKC submitted 2025-11-20 stat.ML cs.LGmath.SP

classification stat.MLcs.LGmath.SP MSC 94A1205C5015A16
keywords graphfractionalFouriertransformangularrotationmatrixfamilydegeneracy-friendlysignaldenoisingoptimalfilteringunitaryparameteroptimization
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 a graph Fourier transform that carries two dials at once: a fractional order α and a rotation angle θ of the spectral basis. Its central fix is a rotation matrix family that returns exactly the identity at θ=0, so the transform reduces to the plain graph fractional Fourier transform at zero angle—something existing angular graph Fourier transforms fail to do. On top of this degeneracy-friendly rotation family, the paper defines two variants (I-AGFRFT and II-AGFRFT), proves they are unitary, invertible, and smooth in both parameters, and uses the parameters as learnable degrees of freedom in an optimal filtering pipeline. Experiments on time series, images, and point clouds report lower MSE and higher PSNR/SSIM than GFRFT, AGFT, and baseline filters.

What carries the argument

The degeneracy-friendly rotation matrix family in SO(N) is the central object. It is built recursively: for even N=2M the rotation is a product of a block-diagonal combination of the lower-dimensional rotation and its diamond double-flip, a block Givens rotation S_M(ϕ), and an exponential map exp(ϕJ) with axis-dependent skew-symmetric J. The diamond flip and the exponential perturbation make the family reduce to the identity at θ=0 while keeping yaw, pitch, and roll distinct in high dimensions; this exact reduction carries the paper's consistency claims.

What would settle it

Split each denoising experiment into training and test signals: optimize (θ, α, κ, H) on one noisy-clean pair (or one subset) and evaluate on a held-out noisy observation. If the MSE advantage of AGFRFT over GFRFT and AGFT largely disappears on held-out signals, the claimed denoising superiority is an artifact of in-sample parameter fitting.

Watch

Extended reading notes

Core claim

The central claim is that fractional order and angular rotation can be unified in one graph spectral transform without sacrificing theoretical consistency. The key is a recursively built rotation family R(θ) in SO(N) with R(0)=I_N, constructed from block Givens rotations, a 'diamond' double-flip, and axis-dependent skew-symmetric matrices fed through the matrix exponential. This guarantees exact reduction: at θ=0 the AGFRFT becomes GFRFT, and at α=1 it becomes a corrected AGFT. Two variants are defined—Type I as (F_θ)^α and Type II as F^α R(θ)^H—and the paper proves unitarity, reversibility, smooth parameter dependence, and (for Type I) index additivity. The paper then treats θ, α, and the f

Load-bearing premise

The denoising evaluation optimizes the transform parameters and filter with the clean target signal in the loss on the same signals being evaluated, so the reported improvements over baselines are in-sample fitting results rather than demonstrated generalization to unseen noisy signals.

Editorial extensions

If this is right

  • At zero rotation angle, AGFRFT reduces exactly to GFRFT, so any angular rotation is a true generalization rather than a different transform.
  • Type I AGFRFT inherits index additivity, letting fractional orders compose additively at fixed angle; Type II does not unless a commutation condition holds.
  • The transform is unitary and invertible for both variants, preserving energy and allowing exact reconstruction via negative fractional order.
  • The whole pipeline—rotation matrix, fractional matrix power, and filter—is differentiable in (θ, α, κ), enabling gradient-based joint optimization for large graphs.
  • Reported experiments show consistent MSE/PSNR/SSIM gains over GFRFT and AGFT on temporal, image, and point-cloud denoising.

Reading between the lines

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

  • Because the rotation family is built only from identity-preserving blocks, the same construction could be dropped into any GFRFT-based algorithm to add angular control without breaking its zero-angle behavior.
  • The reported superiority is measured by optimizing the filter and parameters on the very signals used for evaluation; a test on held-out noisy signals would be needed to confirm the gains are predictive rather than in-sample fitting.
  • The two variants trade off: Type I has cleaner algebraic structure, while Type II is simpler to implement but needs commutation for index additivity; applications needing fractional-order composition should prefer Type I.
  • The exponential-map perturbation could be explored as a general mechanism for parameterizing smooth families of graph filters with guaranteed identity reduction.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper proposes a graph spectral transform that combines the graph fractional Fourier transform (GFRFT) with a rotation of the graph Fourier basis, called RP-GFRFT in the arXiv title and AGFRFT in the body. The construction uses a recursively defined, degeneracy-friendly rotation family that reduces to the identity at zero angle, plus two variants (Type I and Type II) whose unitarity, reduction, index-additivity, and invertibility properties are stated with proofs in the appendices. The authors then propose grid-search and gradient-descent algorithms to jointly optimize the fractional order, rotation angle, rotation scale, and a Wiener filter by minimizing the MSE between the filtered output and the clean signal. Experiments on time series, images, and point clouds report improved MSE/PSNR/SSIM over GFRFT, AGFT, and other baselines.

Significance. The theoretical construction is plausible and the degeneracy property is a genuine improvement over the existing AGFT construction, provided the proofs in Appendices A–C are accepted. The properties stated (orthogonality, exact reduction at θ=0, unitarity, invertibility) are natural and useful for a parameterized graph spectral transform. However, the paper's significance as stated in the abstract rests on the empirical claim of improved denoising, and that claim is not supported by the reported protocol: the clean signal is used to select the transform parameters and the filter on the same signals used for evaluation. The reported numbers are therefore in-sample fitting results, not predictive denoising performance. The lack of code, data, error bars, and held-out evaluation further weakens the empirical contribution.

major comments (2)
  1. [§IV-A, §IV-B, Eq. (26)-(28), Algorithms 1-2, Tables II-VI] The empirical validation is circular. Algorithm 1 (lines 1, 10) and Algorithm 2 (lines 1, 8) take the clean signal x as an input and select θ, α, κ, H by minimizing ||F^{-1}HF y − x||^2 on the same y that is later evaluated. Eq. (26) and Eq. (28) formalize this in-sample objective. Tables II–VI report MSE/PSNR/SSIM on exactly these fitted signals. Because AGFRFT contains additional free parameters (θ, κ) relative to GFRFT/AGFT, lower in-sample error is a guaranteed artifact of the extra flexibility, not evidence of better denoising. A valid experiment must fix parameters on training data (or use cross-validation) and report performance on held-out noisy signals with multiple noise realizations and error bars.
  2. [§IV-B, Eq. (28), Algorithm 2] As written, the optimization admits a trivial zero-loss solution. For any fixed unitary F=F_{θ,α,κ}, the diagonal choice H_ii=(Fx)_i/(Fy)_i for all i with (Fy)_i≠0 achieves F^{-1}HF y=x exactly. The paper does not restrict H to a parametric Wiener form or otherwise regularize the optimization, so the reported nonzero MSE values in Tables V–VI are inconsistent with the stated objective unless an omitted constraint was used. This under-specification further undermines the experimental section and needs to be clarified.
minor comments (5)
  1. [Title/Abstract/§I] The arXiv title and the supplied abstract refer to 'RP-GFRFT', while the manuscript header, full-text abstract, and body consistently use 'AGFRFT'. This naming inconsistency must be resolved before any revision.
  2. [Appendix B, Theorem 1(4)] The claim that yaw/pitch/roll families remain distinct is asserted from the sparsity patterns of J_axis but not actually proved. Distinct generators do not automatically imply distinct matrix families after multiplication by blkdiag and S_M. Please provide a proof or state this as an explicit assumption.
  3. [Tables II-IV] The tables are extremely dense and use combined cells for multiple σ values, which makes them difficult to read. Consider splitting them or using a clearer layout. Also, Table V contains a suspicious non-monotonicity (GFRFT Parrot σ=30 PSNR 48.330 > σ=20 PSNR 46.183) that should be checked.
  4. [§V] No code or data availability statement is provided, and no error bars or number of noise realizations are reported. This is important for assessing the stability of the claimed improvements.
  5. [§III-B] The definition of J_roll states i,j=1,...,N−1 but the matrix is claimed to be N×N; clarify how the last row and column are defined.

Circularity Check

2 steps flagged · score 6.0 of 10

Denoising results are in-sample fits: Algorithms 1–2 and Eq. (28) use the clean target x to choose θ, α, κ, and H, then Tables II–VI report metrics on the same signals.

  1. fitted input called prediction [Section IV-A, Algorithm 1 and Eq. (26); Section V-A Tables II–IV]
    "Algorithm 1 Grid Search for Optimal AGFRFT Filtering 1: Input:y,x,G, AGFRFT type (I/II) ... 10: MSE← ∥ ex−x∥ 2 2/N ... (26) (θ∗, α∗) = arg min θ,α E n ∥ex(θ, α)−x∥2 2 o"

    The grid search takes the clean target x as an input and selects θ∗, α∗, and H∗ by minimizing the MSE between the filtered noisy signal and x on the same y. Section V then reports MSE/PSNR/SSIM on those same y/x instances (Tables II–IV) with no held-out noisy test set or separate clean target described. The reported 'denoising accuracy' is therefore the training objective itself, and AGFRFT's extra free parameters make a better in-sample fit expected by construction.

  2. fitted input called prediction [Section IV-B, Algorithm 2 and Eq. (28); Section V-B/C Tables V–VI]
    "Algorithm 2 Gradient Descent for AGFRFT Filtering 1: Input:y,x,G ... 8: L (t) ← ∥ex(t) −x∥ 2 2 (Eq. 28) ... 16: return (θ ∗, α∗, κ∗),H ∗ ; Eq. (28): L(H,θ,α,κ)=∥F −1 θ,α,κHFθ,α,κy−x∥ 2 2"

    Gradient descent jointly optimizes H, θ, α, and κ against L = ||F^{-1} H F y − x||² using the very clean target x whose reconstruction is later reported, and the final parameters are selected by tracking the lowest training loss L^{(t)}. Because Section V-B/C evaluates PSNR/MSE/SSIM on the same y and x used for this optimization, the claimed 'improvements' over GFRFT and AGFT are in-sample fitting gains, not predictive denoising results.

full rationale

The theoretical core of the paper—the degeneracy-friendly rotation family R^axis_N(θ), the I/II-AGFRFT definitions, and Properties 1–5 (identity, reduction, index additivity, unitarity, reversibility)—is a self-contained mathematical construction. Its proofs do not depend on the experiments or on prior work by the authors; the reduction at θ=0 is built into the definition of the rotation family rather than imported as an external uniqueness result. No self-citation chain is load-bearing here. The circularity is confined to the empirical validation: Algorithms 1 and 2 explicitly take y and the clean signal x as inputs and select θ, α, κ, and H by minimizing ||F^{-1} H F y − x||² on the same instances that Tables II–VI score. In-sample minimization of this loss cannot support the abstract's claim that RP-GFRFT/AGFRFT 'improves denoising accuracy, reconstruction quality, and feature preservation,' because extra tunable parameters guarantee at least as low a training MSE as the less flexible baselines. A held-out noisy test signal, or a parameter-free construction with stated noise assumptions, would be needed to make the empirical claim non-circular. Hence the overall score is 6: partial circularity in the reported predictions, while the transform definition itself is independent.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The transform rests on standard spectral graph theory and matrix-exponential facts, plus the new rotation construction. The most consequential addition is not a physical entity but an evaluation protocol: the clean target x is used to fit all tunable parameters, which makes the empirical claims circular. The free parameters α, θ, κ, and H are all fitted to the evaluation data, and the only truly new mathematical objects—the J matrices and rotation family—have no independent evidence beyond this paper.

free parameters (4)
  • fractional order α = grid [0.1, 1.0] step 0.1 for time series; gradient descent for images/point clouds
    Optimized to minimize MSE against the clean target x in all experiments; central to the transform and to the reported performance.
  • rotation angle θ = grid [0, 2π] step 0.628 for time series; gradient descent for images/point clouds
    Optimized jointly with α; the experiments report specific θ values per dataset and condition.
  • rotation scale κ in ϕ(θ)=κθ = κ=1 for grid search; optimized by gradient descent in Algorithm 2
    Introduced ad hoc to control the angle mapping; affects the rotation family and is not derived from the problem.
  • Wiener filter coefficients H = closed-form Wiener-Hopf solution or gradient descent, always using clean target x
    Not a transform parameter but part of the reported denoising pipeline; fitted directly to the signal being evaluated.
assumptions (5)
  • standard math The graph shift operator has an orthonormal eigendecomposition UΛU^H for undirected weighted graphs.
    Section II.B, Eq. (1); standard for symmetric/normal graph shift operators.
  • domain assumption The matrix power A^α := exp(α log A) via the principal logarithm is well-defined and unitary for unitary A.
    Section III.D and Property 4; depends on a branch choice, and unitarity can fail if eigenvalues hit the branch cut. The paper acknowledges 'appropriate choice of logarithm branch' but does not characterize when such a choice exists for arbitrary graphs.
  • domain assumption The recursively constructed rotation matrices R_N^axis(θ) lie in SO(N) and satisfy R(0)=I.
    Theorem 1; the proof relies on standard properties of matrix exponentials and block products, but the construction itself is new and not independently verified beyond the appendix.
  • ad hoc to paper Optimizing (θ, α, κ, H) by minimizing the MSE between the filtered output and the clean signal x is a valid way to evaluate denoising performance.
    Algorithms 1 and 2 take x as input and minimize ||F^{-1}HFy - x||^2. This makes the reported results in-sample fitting, not a predictive evaluation.
  • domain assumption The k-NN graphs constructed for image blocks (4-NN) and point cloud patches (10-NN) capture the relevant structure for denoising.
    Section V; no sensitivity analysis or comparison to alternative graph constructions is provided.
invented entities (2)
  • Axis-dependent skew-symmetric matrices J_roll, J_pitch, J_yaw
    purpose: Distinguish yaw/pitch/roll rotation families in high dimensions via different sparsity patterns.
    Introduced specifically for this construction; no external validation that these sparsity patterns are the physically or spectrally correct high-dimensional rotations.
  • Degeneracy-friendly rotation matrix family R_N^axis(θ)
    purpose: Provide angular rotation of GFT eigenvectors that reduces to the identity at θ=0.
    New construction used throughout the paper; no independent benchmarks or applications outside this work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rotation-Parameterized Graph Fractional Fourier Transform: Definition, Properties, and Optimal Filtering." pith.science (2026). https://pith.science/paper/RYZI4TKC

@misc{pith2026251116111,
  author       = {Pith},
  title        = {Pith review of: Rotation-Parameterized Graph Fractional Fourier Transform: Definition, Properties, and Optimal Filtering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RYZI4TKC}},
  note         = {Machine review of arXiv:2511.16111}
}
read the original abstract

Graph spectral representations are fundamental in graph signal processing, providing a rigorous frameworkforanalyzing graph-structured data. The graph fractional Fourier transform (GFRFT) extends the graph Fourier transform (GFT) through a fractional-order parameter, enabling flexible spectral analysis with mathematical consistency. The angular graph Fourier transform (AGFT) further introduces angular control by rotating GFT eigenvectors; however, existing constructions may fail to reduce exactly to the GFT at zero angle, weakening theoretical consistency and interpretability. To address these complementary limitations, namely the lack of rotation-based basis control in GFRFT and the defective zero-angle degeneracy of AGFT, this paper proposes the rotation-parameterized graph fractional Fourier transform (RP-GFRFT), which unifies fractional order and rotation-parameterized spectral analysis. A degeneracy preserving rotation matrix family is constructed to guarantee exact GFT reduction at zero angle. TwoRP-GFRFTvariants,I-RP-GFRFTandII-RP-GFRFT,arethenformulated, with theoretical analyses confirming their unitarity, invertibility, reduction behavior, and smooth parameter dependence. The fractional order and rotation angle are jointly optimized for adaptive graph spectral filtering. Experiments on real-world signals, images, and point clouds demonstrate that RP-GFRFT improves denoising accuracy, reconstruction quality, and feature preservation over GFRFT, AGFT, and representative filtering baselines.

Figures

Figures reproduced from arXiv: 2511.16111 by the authors.

Figure 1
Figure 1. Overview of the proposed AGFRFT framework and its applications. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Denoising results on Peppers, Starfish, Monarch, and Parrot images ( [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Denoising results on Sarah9, Ricardo9, Andrew9, and David9 (σ = 40). Columns: Original, Noisy, GFRFT, AGFT, I-AGFRFT, II-AGFRFT. SSIM of 0.9980, significantly exceeding GFRFT (46.449 dB, 2.265e-05, 0.9915) and the best-performing AGFT variant (AGFT with yaw rotation: 39.617 dB, 1.092e-04, 0.9633). Even at a higher noise level(σ = 40), II-AGFRFT with roll rotation maintains superior performance on Peppers, achieving … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 3 linked inside Pith

  1. [1]

    Discrete signal processing on graphs,

    A. Sandryhaila and J. M. Moura, “Discrete signal processing on graphs,” IEEE Trans. Signal Process., vol. 61, no. 7, pp. 1644–1656, Apr. 2013

  2. [2]

    Graph signal processing approach to qsar/qspr model learning of compounds,

    X. Song, L. Chai, and J. Zhang, “Graph signal processing approach to qsar/qspr model learning of compounds,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 4, pp. 1963–1973, Apr. 2020

  3. [3]

    Graph signal processing: Overview, challenges, and applications,

    A. Ortega, P. Frossard, J. Kova ˇcevi´c, J. M. Moura, and P. Vandergheynst, “Graph signal processing: Overview, challenges, and applications,”Proc. IEEE, vol. 106, no. 5, pp. 808–828, May 2018

  4. [4]

    Big data analysis with signal processing on graphs: Representation and processing of massive data sets with irregular structure,

    A. Sandryhaila and J. M. Moura, “Big data analysis with signal processing on graphs: Representation and processing of massive data sets with irregular structure,”IEEE Signal Process. Mag., vol. 31, no. 5, pp. 80–90, Sep. 2014

  5. [5]

    Graph signal processing–part i: Graphs, graph spectra, and spectral clustering,

    L. Stankovic, D. Mandic, M. Dakovic, M. Brajovic, B. Scalzo, and T. Constantinides, “Graph signal processing–part i: Graphs, graph spectra, and spectral clustering,”arXiv preprint arXiv:1907.03467, 2019

  6. [6]

    Graph signal processing–part ii: Processing and analyzing signals on graphs,

    L. Stankovic, D. Mandic, M. Dakovic, M. Brajovic, B. Scalzo, and A. G. Constantinides, “Graph signal processing–part ii: Processing and analyzing signals on graphs,”arXiv preprint arXiv:1909.10325, 2019

  7. [7]

    Graph signal processing, graph neural network and graph learning on biological data: A systematic review,

    R. Li, X. Yuan, M. Radfar, P. Marendy, W. Ni, T. J. O’Brien, and P. M. Casillas-Espinosa, “Graph signal processing, graph neural network and graph learning on biological data: A systematic review,”IEEE Rev. Biomed. Eng., vol. 16, pp. 109–135, 2021

  8. [8]

    Graph signal processing: History, development, impact, and outlook,

    G. Leus, A. G. Marques, J. M. Moura, A. Ortega, and D. I. Shuman, “Graph signal processing: History, development, impact, and outlook,” IEEE Signal Process. Mag., vol. 40, no. 4, pp. 49–60, Jul. 2023

Show all 53 references
  1. [9]

    Graph filters for signal processing and machine learning on graphs,

    E. Isufi, F. Gama, D. I. Shuman, and S. Segarra, “Graph filters for signal processing and machine learning on graphs,”IEEE Trans. Signal Process., vol. 72, pp. 4745–4781, 2024

  2. [10]

    Vertex-frequency graph signal processing: A comprehensive review,

    L. Stankovi ´c, D. Mandic, M. Dakovi´c, B. Scalzo, M. Brajovi´c, E. Sejdi´c, and A. G. Constantinides, “Vertex-frequency graph signal processing: A comprehensive review,”Digit. Signal Process., vol. 107, p. 102802, 2020

  3. [11]

    Learning parametric dictionaries for signals on graphs,

    D. Thanou, D. I. Shuman, and P. Frossard, “Learning parametric dictionaries for signals on graphs,”IEEE Trans. Signal Process., vol. 62, no. 15, pp. 3849–3862, 2014

  4. [12]

    Signal representations on graphs: Tools and applications,

    S. Chen, R. Varma, A. Singh, and J. Kova ˇcevi´c, “Signal representations on graphs: Tools and applications,”arXiv preprint arXiv:1512.05406, 2015. 13

  5. [13]

    Complex graph analysis and representation learning: Problems, techniques, and applications,

    X. Pei, X. Deng, N. N. Xiong, S. Mumtaz, and J. Wu, “Complex graph analysis and representation learning: Problems, techniques, and applications,”IEEE Trans. Netw. Sci. Eng., vol. 11, no. 5, pp. 4990– 5007, 2024

  6. [14]

    Graph spectral image processing,

    G. Cheung, E. Magli, Y . Tanaka, and M. K. Ng, “Graph spectral image processing,”Proc. IEEE, vol. 106, no. 5, pp. 907–930, May 2018

  7. [15]

    Ortega,Introduction to Graph Signal Processing

    A. Ortega,Introduction to Graph Signal Processing. Cambridge Univ. Press, 2022

  8. [16]

    Translation on graphs: An isometric shift operator,

    B. Girault, P. Gonc ¸alves, and ´E. Fleury, “Translation on graphs: An isometric shift operator,”IEEE Signal Process. Lett., vol. 22, no. 12, pp. 2416–2420, Dec. 2015

  9. [17]

    Semi-supervised multiresolution classification using adaptive graph filtering with application to indirect bridge structural health monitoring,

    S. Chen, F. Cerda, P. Rizzo, J. Bielak, J. H. Garrett, and J. Kova ˇcevi´c, “Semi-supervised multiresolution classification using adaptive graph filtering with application to indirect bridge structural health monitoring,” IEEE Trans. Signal Process., vol. 62, no. 11, pp. 2879–...

  10. [18]

    Verifying the smoothness of graph signals: A graph signal processing approach,

    L. Dabush and T. Routtenberg, “Verifying the smoothness of graph signals: A graph signal processing approach,”IEEE Trans. Signal Process., 2024

  11. [19]

    Fast graph Fourier transforms based on graph symmetry and bipartition,

    K. Lu and A. Ortega, “Fast graph Fourier transforms based on graph symmetry and bipartition,”IEEE Trans. Signal Process., vol. 67, no. 18, pp. 4855–4869, Sep. 2019

  12. [20]

    F. R. K. Chung,Spectral Graph Theory. Amer. Math. Soc., 1997, vol. 92

  13. [21]

    A matrix representation of graphs and its spectrum as a graph invariant,

    D. Emms, E. R. Hancock, S. Severini, and R. C. Wilson, “A matrix representation of graphs and its spectrum as a graph invariant,”Electron. J. Combin., vol. 13, no. 1, pp. 1–14, 2006

  14. [22]

    Signless Laplacians of finite graphs,

    D. Cvetkovi ´c, P. Rowlinson, and S. K. Simi ´c, “Signless Laplacians of finite graphs,”Linear Algebra Appl., vol. 423, no. 1, pp. 155–171, 2007

  15. [23]

    Signless Laplacian spectrum of a graph,

    A. H. Ghodrati and M. A. Hosseinzadeh, “Signless Laplacian spectrum of a graph,”Linear Algebra Appl., vol. 682, pp. 257–267, 2024

  16. [24]

    Graph signal processing over a probability space of shift operators,

    F. Ji, W. P. Tay, and A. Ortega, “Graph signal processing over a probability space of shift operators,”IEEE Trans. Signal Process., vol. 71, pp. 1159–1174, 2023

  17. [25]

    Fractional graph spectral filtering based on unified graph representation matrix,

    F. Zhao and Z. Zhang, “Fractional graph spectral filtering based on unified graph representation matrix,”IEEE Signal Process. Lett., 2025

  18. [26]

    The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains,

    D. I. Shuman, S. K. Narang, P. Frossard, A. Ortega, and P. Vandergheynst, “The emerging field of signal processing on graphs: Extending high-dimensional data analysis to networks and other irregular domains,”IEEE Signal Process. Mag., vol. 30, no. 3, pp. 83–98, May 2013

  19. [27]

    Discrete signal processing on graphs: Graph Fourier transform,

    A. Sandryhaila and J. M. Moura, “Discrete signal processing on graphs: Graph Fourier transform,” inProc. IEEE Int. Conf. Acoust., Speech, Signal Process., 2013, pp. 6167–6170

  20. [28]

    Graph signal processing for directed graphs based on the Hermitian Laplacian,

    S. Furutani, T. Shibahara, M. Akiyama, K. Hato, and M. Aida, “Graph signal processing for directed graphs based on the Hermitian Laplacian,” inProc. Mach. Learn. Knowl. Discov. Databases. U. Brefeld, E. Fromont, A. Hotho, A. Knobbe, M. Maathuis, and C. Robardet, Eds., Cham, Sw...

  21. [29]

    Discrete signal processing on graphs: Frequency analysis,

    A. Sandryhaila and J. M. Moura, “Discrete signal processing on graphs: Frequency analysis,”IEEE Trans. Signal Process., vol. 62, no. 12, pp. 3042–3054, Jun. 2014

  22. [30]

    A directed graph Fourier transform with spread frequency components,

    R. Shafipour, A. Khodabakhsh, G. Mateos, and E. Nikolova, “A directed graph Fourier transform with spread frequency components,”IEEE Trans. Signal Process., vol. 67, no. 4, pp. 946–960, Feb. 2019

  23. [31]

    On the graph Fourier transform for directed graphs,

    S. Sardellitti, S. Barbarossa, and P. Di Lorenzo, “On the graph Fourier transform for directed graphs,”IEEE J. Sel. Topics Signal Process., vol. 11, no. 6, pp. 796–811, Sep. 2017

  24. [32]

    Discrete signal processing on graphs: Graph filters,

    A. Sandryhaila and J. M. Moura, “Discrete signal processing on graphs: Graph filters,” inProc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP). IEEE, May 2013, pp. 6163–6166

  25. [33]

    Graph Fourier transform: A stable approximation,

    J. Domingos and J. M. Moura, “Graph Fourier transform: A stable approximation,”IEEE Trans. Signal Process., vol. 68, pp. 4422–4437, 2020

  26. [34]

    The fractional Fourier transform on graphs,

    Y . Wang, B. Li, and Q. Cheng, “The fractional Fourier transform on graphs,” inProc. Asia-Pacific Signal Inf. Process. Assoc. Annu. Summit Conf., Dec. 2017, pp. 105–110

  27. [35]

    The fractional Fourier transform on graphs: Sampling and recovery,

    Y . Wang and B. Li, “The fractional Fourier transform on graphs: Sampling and recovery,” inProc. 14th IEEE Int. Conf. Signal Process., 2018, pp. 1103–1108

  28. [36]

    Generalized sampling of graph signals with prior information based on graph fractional Fourier transform,

    D. Wei and Z. Yan, “Generalized sampling of graph signals with prior information based on graph fractional Fourier transform,”Signal Process., vol. 214, p. 109263, 2024

  29. [37]

    Graph fractional Fourier transform: A unified theory,

    T. Alikas ¸ifo ˘glu, B. Kartal, and A. Koc ¸, “Graph fractional Fourier transform: A unified theory,”IEEE Trans. Signal Process., vol. 72, pp. 3834–3850, 2024

  30. [38]

    Multi-dimensional graph fractional Fourier transform and its application to data compression,

    F. Yan and B. Li, “Multi-dimensional graph fractional Fourier transform and its application to data compression,”Digit. Signal Process., vol. 129, p. 103683, 2022

  31. [39]

    Trainable fractional Fourier transform,

    E. Koc ¸, T. Alikas ¸ifo˘glu, A. C. Aras, and A. Koc ¸, “Trainable fractional Fourier transform,”IEEE Signal Process. Lett., vol. 31, pp. 751–755, 2024

  32. [40]

    Spectral graph fractional Fourier transform for directed graphs and its application,

    F. Yan and B. Li, “Spectral graph fractional Fourier transform for directed graphs and its application,”Signal Process., vol. 210, p. 109099, 2023

  33. [41]

    Joint time-vertex fractional Fourier transform,

    T. Alikas ¸ifo˘glu, B. Kartal, E. ¨Ozg¨unay, and A. Koc ¸, “Joint time-vertex fractional Fourier transform,”Signal Process., vol. 233, p. 109944, 2025

  34. [42]

    Windowed fractional Fourier transform on graphs: Properties and fast algorithm,

    F. Yan and B. Li, “Windowed fractional Fourier transform on graphs: Properties and fast algorithm,”Digit. Signal Process., vol. 118, p. 103210, 2021

  35. [43]

    Color image encryption using angular graph Fourier transform,

    L. Yang, W. Meng, and X. Zhao, “Color image encryption using angular graph Fourier transform,”Int. J. Digit. Crime Forensics, vol. 13, no. 3, pp. 59–82, 2021

  36. [44]

    On the shift operator, graph frequency, and optimal filtering in graph signal processing,

    A. Gavili and X. Zhang, “On the shift operator, graph frequency, and optimal filtering in graph signal processing,”IEEE Trans. Signal Process., vol. 65, no. 23, pp. 6303–6318, Dec. 2017

  37. [45]

    Algorithms for decomposing 3-d orthogonal matrices into primitive rotations,

    T. Herter and K. Lott, “Algorithms for decomposing 3-d orthogonal matrices into primitive rotations,”Comput. Graph., vol. 17, no. 5, pp. 517–527, 1993

  38. [46]

    Rotation matrix,

    E. W. Weisstein, “Rotation matrix,”MathWorld–A Wolfram Web Resource, 2003

  39. [47]

    The mathematics of the 3d rotation matrix,

    D. Gruber, “The mathematics of the 3d rotation matrix,” inProc. Xtreme Game Developers Conf., 2000, pp. 1–14

  40. [48]

    Metrics for 3d rotations: Comparison and analysis,

    D. Q. Huynh, “Metrics for 3d rotations: Comparison and analysis,”J. Math. Imaging Vis., vol. 35, no. 2, pp. 155–164, 2009

  41. [49]

    Optimal filtering in fractional Fourier domains,

    A. Kutay, H. M. Ozaktas, O. Ankan, and L. Onural, “Optimal filtering in fractional Fourier domains,”IEEE Trans. Signal Process., vol. 45, no. 5, pp. 1129–1143, May 1997

  42. [50]

    Optimal fractional Fourier filtering for graph signals,

    C. Ozturk, H. M. Ozaktas, S. Gezici, and A. Koc ¸, “Optimal fractional Fourier filtering for graph signals,”IEEE Trans. Signal Process., vol. 69, pp. 2902–2912, May 2021

  43. [51]

    The optimal joint time-vertex graph filter design: From ordinary graph Fourier domains to fractional graph Fourier domains,

    Z. Ge, H. Guo, T. Wang, and Z. Yang, “The optimal joint time-vertex graph filter design: From ordinary graph Fourier domains to fractional graph Fourier domains,”Circuits, Syst., Signal Process., vol. 42, no. 7, pp. 4002–4018, 2023

  44. [52]

    USC-SIPI Image Database, https://sipi.usc.edu/database/

  45. [53]

    Jpeg pleno database: microsoft voxelized upper bodies-a voxelized point cloud dataset,

    L. Charles, Q. Cai, E. S. Orts, and A. C. Philip, “Jpeg pleno database: microsoft voxelized upper bodies-a voxelized point cloud dataset,” 2021

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.