Pith. sign in

REVIEW 5 major objections 6 minor 36 references

Color Image Set Recognition Based on Quaternionic Grassmannians

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a closed-form shortest-geodesic distance on quaternionic Grassmannians, combined with a nearest-centroid rule, recognizes color image sets with high accuracy.

desk verdict A fresh application idea with a working implementation, but the central distance formula is not proven and the metric proof contains an apparent false identity. read the letter →

arxiv 2505.23629 v2 pith:ZCEH7N65 submitted 2025-05-29 cs.CV math.AG

Xiang Xiang Wang , Tin-Yau Tam This is my paper
classification cs.CVmath.AG
keywords quaternionicGrassmanniancolorimagesetrecognitionshortestgeodesicdistancestandardeigenvaluesquaternionPCAnearest-centroidclassificationETH-80HighwayTrafficdataset
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

This paper claims that color image sets can be recognized by lifting each set to a point on a quaternionic Grassmannian and using a newly derived closed-form formula for the shortest geodesic distance between two such points. The central contribution is Theorem 3.4, which expresses that distance as half the root-sum-square of the arguments of the standard eigenvalues of the quaternionic unitary matrix $(I-2Q)(I-2P)$. The authors argue this quantity is a genuine metric and build a nearest-centroid classifier on it. On the ETH-80 benchmark the classifier reaches 97.00% average accuracy, and on the Highway Traffic video dataset it reaches 88.55%, both above the baselines compared in the paper.

What carries the argument

The load-bearing object is the quaternionic Grassmannian $\mathrm{Gr}_{n,k}(\mathbb{H})$, realized as the set of $n\times n$ quaternionic orthogonal projection matrices of rank $k$. Geodesics are written $\gamma(t)=\exp(tX)P\exp(-tX)$ with $X$ skew-Hermitian quaternionic, and the endpoint condition $\exp(2X)=(I-2Q)(I-2P)$ links the geodesic generator to the two endpoints. The distance formula then reads off the generator's size from the arguments of the standard eigenvalues of that unitary product. Color enters by encoding each RGB image as a pure quaternionic matrix, reducing the set with quaternionic PCA, orthonormalizing, and forming the projection $XX^\ast$.

What would settle it

Take two points $P,Q$ in $\mathrm{Gr}_{n,k}(\mathbb{H})$ such that $(I-2Q)(I-2P)$ has $-1$ as an eigenvalue and compute the formula's value under both choices $\arg(-1)=\pm\pi$; if the two values differ while a numerical geodesic-distance optimizer gives a single number, the formula as stated is not well-defined. Alternatively, randomly sample triples with eigenvalues near $-1$ and check the triangle inequality numerically; a violation would refute the metric claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that for any two points $P$ and $Q$ on the quaternionic Grassmannian $\mathrm{Gr}_{n,k}(\mathbb{H})$, the shortest geodesic distance is $$ \hat d(P,Q)=\frac12\sqrt{\sum_j \$arg^{2}$\big(\hat\lambda_j((I-2Q)(I-2P))\big)}, $$ where $\hat\lambda_j$ runs over the standard eigenvalues of the quaternionic unitary matrix. The authors state this formula is compatible with real and complex Grassmannians, that it satisfies non-negativity, symmetry, and the triangle inequality, and that it supplies the distance used in their classification framework. The proof routes through the geodesic equation $\gamma(t)=e^{tX}Pe^{-tX}$ with boundary condition $e^{2X}=(I-2Q)(I-2P)$, the quaternionic Schatten-2 norm, and the complex representation of quaternionic matrices.

Load-bearing premise

The proof assumes that the shortest geodesic is obtained by counting each principal angle exactly once, i.e., that the norm of the optimal generator equals the square root of the sum of squared arguments of the standard eigenvalues, with no extra multiplicities and no inconsistency in choosing the argument of $-1$; if that step is false, the formula is only a partial or scaled distance.

Editorial extensions

If this is right

  • On the paper's terms, the formula supplies a metric on $\mathrm{Gr}_{n,k}(\mathbb{H})$, so nearest-centroid and nearest-neighbor classifiers can be run directly on color image sets without splitting RGB channels.
  • The framework extends to color video by treating each video as a set of frames and mapping the set to one quaternionic Grassmannian point.
  • Because the formula is claimed compatible with real and complex Grassmannians, the same distance expression can serve subspace comparisons in settings beyond color.
  • The reported ETH-80 accuracy of 97.00% and Highway Traffic accuracy of 88.55% indicate the representation preserves enough chromatic and geometric structure to beat the compared subspace baselines.
  • The computation of standard eigenvalues of $(I-2Q)(I-2P)$ is the main bottleneck, so faster eigenvalue algorithms would make the method practical for large image sets.

Reading between the lines

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

  • If the distance formula is genuinely a metric, a natural extension is to feed it into discriminative learning on quaternionic Grassmannians; the paper itself only uses a nearest-centroid rule.
  • The branch ambiguity at eigenvalue $-1$ is the spot where the argument of the formula could break; a numerical experiment comparing the formula against a direct geodesic optimization on pairs with an eigenvalue exactly at $-1$ would settle whether the claimed shortest distance is correct.
  • The same eigenvalue-argument construction may transfer to other symmetric spaces of unitary type, giving closed-form distances in settings the paper does not discuss.
  • The reported standard deviations (up to about $\pm3.5\%$ across rounds on ETH-80) suggest accuracy depends on the random split, so quantifying how much of the advantage comes from the metric versus the quaternionic representation would clarify the method's real margin.
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

5 major / 6 minor

Summary. The paper proposes representing color image sets as points on quaternionic Grassmannians, where each RGB image is encoded as a quaternionic matrix and a set is summarized by a quaternionic subspace projection matrix. The central theoretical contribution is a claimed closed-form formula (3.9) for the shortest geodesic distance between two points on Gr_{n,k}(H), expressed in terms of the arguments of the standard eigenvalues of (I-2Q)(I-2P). The paper further claims this distance is a metric and uses it in a nearest-centroid classifier evaluated on the ETH-80 dataset and a Highway Traffic video dataset, reporting recognition rates of 97.00% and 88.55%, respectively. The mathematical core is Theorem 3.4 and the metric proof in Section 3.3; the experimental framework in Section 4 and the results in Section 5 depend directly on the unproven distance formula.

Significance. If rigorously established, the quaternionic Grassmannian distance formula would provide a natural color-preserving generalization of complex Grassmannian geodesic distances and a compact representation for image-set recognition that captures inter-channel correlations. The paper has the virtue of shipping MATLAB code and using a simple, reproducible experimental protocol. However, the central proofs contain substantial gaps: the minimization over geodesics in Theorem 3.4 is not derived, the triangle inequality proof in Section 3.3 is circular, and the extension to quaternions in Corollary 3.10 is asserted without proof. Because the experimental results inherit the unproven distance formula, the paper's main claims are not currently supported.

major comments (5)
  1. [§3.2, Theorem 3.4, Eq. (3.10)] The proof's decisive step, 'Since similar eigenvalues have the same norm, we obtain min_j ||2X_j||_H = sqrt(sum_j a_j^2),' is not a consequence of eigenvalue similarity. The Frobenius norm of a skew-Hermitian matrix is determined by the eigenvalues of that matrix, which are only determined up to branch choices of the form i(a_j + 2π m_j). The proof does not show that the principal branch m_j = 0 yields a valid geodesic connecting P and Q, nor does it handle multiplicities arising from passing from the n standard eigenvalues of the quaternionic matrix to the 2n eigenvalues of its complex representation. Without this minimization, formula (3.9) is not established as the shortest distance.
  2. [§3.3, Eq. (3.17)] The identity e^{X}e^{Y}e^{Z} = e^{-X}e^{-Y}e^{-Z} is asserted without proof. The preceding four-term sinh/cosh expression is algebraically equivalent to this identity, not a consequence of (3.16), so the proof of Theorem 3.9 is circular. Furthermore, the derivation of e^{-2X} = e^{Y}e^{Z}e^{Z}e^{Y} from (3.17) appears to treat exponentials of non-commuting matrices as if they could be freely reordered, which is not justified.
  3. [§3.3, Corollary 3.10] The extension of the triangle inequality to quaternionic Grassmannians is asserted to follow from Proposition 2.1 and a reference to [29], but no proof is supplied. Thompson's theorem (Theorem 3.8) is stated for complex skew-Hermitian matrices; its quaternionic analogue requires a nontrivial argument concerning eigenvalues, similarity classes, and the branch of the logarithm, none of which is given. Since the metric property of (3.9) is a central claim, this gap is load-bearing.
  4. [§3.2, Theorem 3.4] The proof does not address the ambiguity of arg(-1) (the eigenvalue -1 can be assigned argument π or -π) or the fact that, per Definition 2.2 and Corollary 2.4, the standard eigenvalues of a quaternionic unitary matrix lie in the closed upper half-plane. The proof writes a_j ∈ [-π,π], which is inconsistent with the stated definition of standard eigenvalues, and the choice of branch is directly relevant to the minimization over geodesics.
  5. [§5, Tables 2 and 3] The Highway Traffic baselines in Table 3 are reported without standard deviations, so the claim that the proposed method 'significantly outperforms' all baselines is not supported; given the reported ±2.46% for the proposed method, overlap with baselines is possible. Additionally, Remark 5.1 shows ETH-80 average rates varying across rounds from 93.0% to 97.25%, so the single reported value 97.00% in Table 2 is not presented as an aggregate over the rounds described, weakening the empirical support.
minor comments (6)
  1. [Abstract] The abstract contains a duplicated word: 'ETH-80 benchmark dataset and and the Highway Traffic video dataset'.
  2. [§4.1, Algorithm 1] Algorithm 1 line 6 says 'Reduce ... by QSVD' while the surrounding text in §4.1 describes the step as Quaternion Principal Component Analysis (QPCA); please unify the terminology.
  3. [§3.2, before Theorem 3.4] The sentence 'Let X ∈ UH(n)' appears to be a typo: the generator of a geodesic on the Grassmannian should be a skew-Hermitian matrix in sH(n), not a unitary matrix.
  4. [§3.3] There is an unresolved citation '[ ?]' in the sentence 'particularly in [ ?] and [27]'.
  5. [§3.3, Proposition 3.7] The notation \(\widetilde{X}_{PQ}\) is used without a formal definition, and the placement of the tilde is inconsistent across the proof of Theorem 3.9.
  6. [§5.1] It is unclear whether the baseline methods in Table 2 were re-run under the same random splits as the proposed method or taken from their original papers; please clarify the comparison protocol.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the central distance formula rests on external results, and the single self-citation [30] is confined to a non-load-bearing remark.

full rationale

Walking the derivation chain: Theorem 3.4's distance formula is built from the geodesic characterization in Remark 3.2 and equation (3.6), attributed to Batzies et al. [27]; the quaternionic SVD and Schatten-2 norm from Zhang [21] and Miao/Kou [28]; and the unitary logarithm. The proof's key minimization step, 'Since similar eigenvalues have the same norm, we obtain min_j ||2X_j||_H = sqrt(sum_j a_j^2),' is asserted without deriving the branch-of-logarithm choice and eigenvalue multiplicities; that is an omitted proof or correctness gap, but it is not a circular reduction to the paper's own inputs. The only self-citation, [30], appears in Remark 3.11 as an optional extension after Corollary 3.10 has already completed the metric proof using Thompson's external theorem, so it is not load-bearing. The experiments use a nearest-centroid classifier on the proposed distance with QPCA representation; no parameter is fitted to test labels or to the reported accuracy, so the 97.00% and 88.55% results are not forced by construction. The missing reference '[ ?]' in Section 3.3 is a presentation flaw, and the stability limitations in Remark 5.1 and the Conclusion are candid but do not affect circularity. The formula has antecedents in the real and complex Grassmannian literature, so novelty may be limited, but limited novelty is not circularity. No self-definition, fitted-input-as-prediction, or author-imported uniqueness argument is present.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim loads on standard quaternion spectral theory and an imported geodesic boundary condition. The only hand-tuned numbers are the subspace dimension and image resizing.

free parameters (2)
  • subspace dimension k = 9
    Set to 9 for both ETH-80 (20x20 images) and Highway Traffic (24x24 frames); no ablation or criterion for choosing 9 is given.
  • image resize dimensions = 20x20 and 24x24
    ETH-80 images resized to 20x20 and Highway frames to 24x24 following prior work; these choices affect dimensionality but are not varied.
assumptions (5)
  • standard math Properties of the complex representation of quaternionic matrices (Proposition 2.1)
    Quoted from Lee 1948 [19]; used to extend complex Grassmannian results to quaternions.
  • standard math Standard eigenvalues of a quaternionic matrix exist with non-negative imaginary part (Definition 2.2)
    From Brenner/Lee, used in Theorem 3.4 to take arguments of eigenvalues.
  • standard math Boundary condition exp(2X) = (I-2Q)(I-2P) for geodesics on the Grassmannian
    From Batzies et al. [27]; the paper does not prove it, it is a load-bearing imported result.
  • domain assumption Thompson's theorem on exponentials of skew-Hermitian matrices and its extension to quaternions
    Theorem 3.8 quoted for complex matrices; the paper asserts without proof that it holds for quaternionic matrices, which is needed for Corollary 3.10.
  • domain assumption The geodesic distance on Gr_{n,k}(H) is the Frobenius norm of the minimal generator X
    The paper assumes the shortest geodesic length equals ||X||_H where X satisfies the boundary condition; the minimization over all geodesics is not fully justified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Color Image Set Recognition Based on Quaternionic Grassmannians." pith.science (2026). https://pith.science/paper/ZCEH7N65

@misc{pith2026250523629,
  author       = {Pith},
  title        = {Pith review of: Color Image Set Recognition Based on Quaternionic Grassmannians},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZCEH7N65}},
  note         = {Machine review of arXiv:2505.23629}
}
read the original abstract

We propose a new method for recognizing color image sets using quaternionic Grassmannians, which use the power of quaternions to capture color information and represent each color image set as a point on the quaternionic Grassmannian. We provide a direct formula to calculate the shortest distance between two points on the quaternionic Grassmannian, and use this distance to build a new classification framework. Experiments on the ETH-80 benchmark dataset and and the Highway Traffic video dataset show that our method achieves good recognition results. We also discuss some limitations in stability and suggest ways the method can be improved in the future.

Figures

Figures reproduced from arXiv: 2505.23629 by the authors.

Figure 1
Figure 1. A color image represented as a pure quaternionic matrix This quaternion-based representation enables compact and unified processing of the color channels while preserving the relationships between them. The key idea is that the quaternionic structure inherently encodes the correlations between the RGB channels, as opposed to treating them as three separate grayscale images. This inter-channel depen￾dency is especial… view at source ↗
Figure 2
Figure 2. Quaternionic Grassmannian representation of a color image set. 4.2. A New Framework for Color Image Set Recognition Based on Quater￾nionic Grassmannians. Based on the Grassmannian representation of image sets and the distance formula, we propose a new framework for color image set recognition using quaternionic Grassmannians. As a straightforward example, we consider three image sets belonging to two distinct classe… view at source ↗
Figure 3
Figure 3. Framework for color image set recognition using quaternionic Grassmannians (three image sets) developing more advanced algorithms and methods in future work, enabling more robust and efficient recognition techniques. 5. Numerical Example In this section, we evaluate the performance of our proposed quaternionic Grassmannian framework on two image set recognition tasks: the ETH-80 dataset [32] and a Highway Traffic da… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The eight categories of the ETH-80 dataset. Each category contains 10 objects with 41 views per object. For this evaluation, the dataset was divided into training and testing sets. Five object instances selected from each category were used for training, and the remain…
Figure 5
Figure 5. Figure 5: Visualization of the eight categories represented by training data points and the selected 10 testing data points. Method Recognition Rate (%) GDA [2] 91.00 ± 2.13 GEDA [3] 92.50 ± 1.16 GDL [8] 93.50 ± 0.92 GiFME [35] 74.50 ± 1.22 GGPLCR [36] 96.75 ± 1.30 Our Method 97…
Figure 6
Figure 6. Figure 6: Sample frames from the Highway Traffic dataset. The dataset includes 44 Heavy, 45 Medium, and 165 Light traffic videos. For the experimental evaluation, we followed the protocol described in Wei et al.’s paper (2024) [11] to ensure a fair comparison. Each video was tre…
Figure 7
Figure 7. Figure 7: Visualization of the three categories represented by training data points and the selected 10 testing data points. Method Recognition Rate (%) GNN [2] 70.00 GLPP [9] 76.67 GDA [2] 75.69 GEDA [3] 78.96 GNPE[10] 75.66 GALL [11] 78.95 F-GALL [11] 79.04 Our Method 88.55 ± …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 34 canonical work pages

  1. [29]

    Princeton Series in Applied Mathematics

    Rodman, L.: Topics in quaternion linear algebra. Princeton Series in Applied Mathematics. Princeton University Press, Princeton (2014)

    work page 2014

  2. [1]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 29(6), 1005– 1018 (2007)

    Kim, T.K., Kittler, J., Cipolla, R.: Discriminative learning and recognition of image set classes using canonical correlations. IEEE Transactions on Pattern Analysis and Machine Intelligence 29(6), 1005– 1018 (2007)

    work page 2007

  3. [2]

    In: Proceedings of the 25th international conference on Machine learning (2008)

    Hamm, J., Lee, D.D.: Grassmann discriminant analysis: a unifying view on subspace-based learning. In: Proceedings of the 25th international conference on Machine learning (2008). https://doi.org/ 10.1145/1390156.1390204

    arXiv 2008

  4. [3]

    Conference on Computer Vision and Pattern Recognition (2011)

    Harandi, M.T., Sanderson, C., Shirazi, S.A., Lovell, B.C.: Graph embedding discriminant analysis on grassmannian manifolds for improved image set matching. Conference on Computer Vision and Pattern Recognition (2011). https://doi.org/10.1109/CVPR.2011.5995564

    work page arXiv 2011

  5. [4]

    In: 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (2003)

    Liu, X., Cheng, T.: Video-based face recognition using adaptive hidden markov models. In: 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (2003). https: //doi.org/10.1109/CVPR.2003.1211373

    arXiv 2003

  6. [5]

    Neurocomputing 517, 20–33 (2023)

    Souza, L.S., Sogi, N., Gatto, B.B., Kobayashi, T., Fukui, K.: Grassmannian learning mutual subspace method for image set recognition. Neurocomputing 517, 20–33 (2023)

    work page 2023

  7. [6]

    IEEE Transactions on Image Processing 20(9), 2627–2635 (2011)

    Wang, X., Li, Z., Tao, D.: Subspaces indexing model on grassmann manifold for image search. IEEE Transactions on Image Processing 20(9), 2627–2635 (2011)

    work page 2011

  8. [7]

    In: Pro- ceedings third IEEE international conference on automatic face and gesture recognition (1998)

    Yamaguchi, O., Fukui, K., Maeda, K.i.: Face recognition using temporal image sequence. In: Pro- ceedings third IEEE international conference on automatic face and gesture recognition (1998). https://doi.org/10.1109/AFGR.1998.670968

    work page arXiv 1998

Show all 36 references
  1. [8]

    IEEE Computer Society (2013).https://doi.org/10.1109/ ICCV.2013.387

    Harandi, M., Sanderson, C., Shen, C., Lovell, B.: Dictionary learning and sparse coding on grass- mann manifolds: An extrinsic solution. IEEE Computer Society (2013).https://doi.org/10.1109/ ICCV.2013.387

    2013

  2. [9]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Kumar, S.: Jumping manifolds: Geometry aware dense non-rigid structure from motion. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 5346–5355 (2019) 23

    2019

  3. [10]

    Pattern Recognition 122, 108335 (2022)

    Wei, D., Shen, X., Sun, Q., Gao, X., Ren, Z.: Neighborhood preserving embedding on grassmann manifold for image-set analysis. Pattern Recognition 122, 108335 (2022)

    2022

  4. [11]

    Expert Systems with Applications 247, 123316 (2024)

    Wei, D., Shen, X., Sun, Q., Gao, X., Ren, Z.: Learning adaptive grassmann neighbors for image-set analysis. Expert Systems with Applications 247, 123316 (2024)

    2024

  5. [12]

    In: Pro- ceedings 2003 International Conference on Image Processing (2003)

    Le Bihan, N., Sangwine, S.J.: Quaternion principal component analysis of color images. In: Pro- ceedings 2003 International Conference on Image Processing (2003). https://doi.org/10.1109/ ICIP.2003.1247085

    2003 arXiv

  6. [13]

    IEEE Transactions on image Processing 8(5), 614–628 (1999)

    Pei, S.C., Cheng, C.M.: Color image processing by using binary quaternion-moment-preserving thresholding technique. IEEE Transactions on image Processing 8(5), 614–628 (1999)

    1999

  7. [14]

    The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 25(163), 10–13 (1844)

    Hamilton, W.R.: On quaternions; or on a new system of imaginaries in algebra. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 25(163), 10–13 (1844)

  8. [15]

    ACM SIGGRAPH Computer Graphics 19(3), 245–254 (1985)

    Shoemake, K.: Animating rotation with quaternion curves. ACM SIGGRAPH Computer Graphics 19(3), 245–254 (1985)

    1985

  9. [16]

    Princeton University Press, Princeton (1999)

    Kuipers, J.B.: Quaternions and Rotation Sequences: A Primer with Applications to Orbits, Aerospace, and Virtual Reality. Princeton University Press, Princeton (1999)

    1999

  10. [17]

    Clarendon Press, Oxford (1986)

    Altmann, S.L.: Rotations, Quaternions, and Double Groups. Clarendon Press, Oxford (1986)

    1986

  11. [18]

    Applied Mathematics and Computation 93(2-3), 195–205 (1998)

    Zhang, F.: Quaternion-based quaternion fourier transform for signal and image processing. Applied Mathematics and Computation 93(2-3), 195–205 (1998)

    1998

  12. [19]

    In: Proceedings of the Royal Irish Academy

    Lee, H.C.: Eigenvalues and canonical forms of matrices with quaternion coefficients. In: Proceedings of the Royal Irish Academy. Section A: Mathematical and Physical Sciences 52, 253–260 (1948)

    1948

  13. [20]

    Pacific Journal of Mathematics 1, 329–335 (1951)

    Brenner, J.L.: Matrices of quaternions. Pacific Journal of Mathematics 1, 329–335 (1951)

    1951

  14. [21]

    Linear algebra and its applications 251, 21–57 (1997)

    Zhang, F.: Quaternions and matrices of quaternions. Linear algebra and its applications 251, 21–57 (1997)

    1997

  15. [22]

    IEEE Transactions on Image Processing 16(1), 22–35 (2007)

    Ell, T.A., Sangwine, S.J.: Hypercomplex fourier transforms of color images. IEEE Transactions on Image Processing 16(1), 22–35 (2007)

    2007

  16. [23]

    Journal of Algebra 59(2), 399–411 (1979)

    Djokovi´ c, D.ˇZ., Malzan, J.: Products of reflections in the quaternionic unitary group. Journal of Algebra 59(2), 399–411 (1979)

    1979

  17. [24]

    IEEE Transactions on Signal Processing 63(6), 1543–1556 (2015)

    Xu, D., Mandic, D.P.: The theory of quaternion matrix derivatives. IEEE Transactions on Signal Processing 63(6), 1543–1556 (2015)

    2015

  18. [25]

    Advances in Computational Mathematics 50, 6 (2024)

    Bendokat, T., Zimmermann, R., Absil, P.A.: A grassmann manifold handbook: Basic geometry and computational aspects. Advances in Computational Mathematics 50, 6 (2024)

    2024

  19. [26]

    Springer, New York (2012)

    Lee, J.M.: Introduction to Smooth Manifolds, Graduate Texts in Mathematics. Springer, New York (2012)

    2012

  20. [27]

    Linear Algebra Application 466, 83–101 (2015)

    Batzies, E., H¨ uper, K., Machado, L., Silva Leite, F.: Geometric mean and geodesic regression on Grassmannians. Linear Algebra Application 466, 83–101 (2015)

    2015

  21. [28]

    IEEE Transactions on Signal Processing 68, 5617–5631 (2020)

    Miao, J., Kou, K.I.: Quaternion-based bilinear factor matrix norm minimization for color image inpainting. IEEE Transactions on Signal Processing 68, 5617–5631 (2020)

    2020

  22. [30]

    arXiv:2412.15161 (2024)

    Tam, T.Y., Wang, X.X.: Geometric properties and distance inequalities on grassmannians. arXiv:2412.15161 (2024)

    2024 arXiv

  23. [31]

    Linear Algebra and its Applications 371, 75–102 (2003)

    Farenick, D.R., Pidkowich, B.A.: The spectral theorem in quaternions. Linear Algebra and its Applications 371, 75–102 (2003)

    2003

  24. [32]

    In: 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (2003)

    Leibe, B., Schiele, B.: Analyzing appearance and contour based methods for object categorization. In: 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (2003). https://doi.org/10.1109/CVPR.2003.1211497

    2003 arXiv

  25. [33]

    In: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition 1, 846–851 (2005)

    Chan, A.B., Vasconcelos, N.: Probabilistic kernels for the classification of auto-regressive visual processes. In: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition 1, 846–851 (2005)

    2005

  26. [34]

    Springer, New York (2005) 24

    Borg, I., Groenen, P.J.F.: Modern Multidimensional Scaling: Theory and Applications. Springer, New York (2005) 24

    2005

  27. [35]

    In: Proceedings of the 2015 IEEE International Conference on Computer Vision (2015)

    Chakraborty, R., Vemuri, B.C.: Recursive fr´ echet mean computation on the grassmannian and its applications to computer vision. In: Proceedings of the 2015 IEEE International Conference on Computer Vision (2015). https://doi.org/10.1109/ICCV.2015.481

    2015 doi

  28. [36]

    Wei, D., Shen, X., Sun, Q., Gao, X., Yan, W.: Prototype learning and collaborative representation using grassmann manifolds for image set classification. Pattern Recognition 100, 107123 (2020) Department of Mathematics and Statistics, University of Nevada, Reno, Reno, NV 89557...

    2020

Pith tools

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