Pith. sign in

REVIEW 4 major objections 8 minor 1 cited by

The paper claims that a narrow Gaussian smoothing scale, σ∈(1,1.2), is necessary and sufficient for second-order Gaussian directional derivative filters to resolve high-resolution adjacent corners in images.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 10:50 UTC pith:BTXD6ZHN

load-bearing objection Useful adjacent-corner SOGDD representations, but the scale-selection derivation is a black box and the σ/σ² slip makes the central claim unverifiable. the 4 major comments →

arxiv 2601.08182 v2 pith:BTXD6ZHN submitted 2026-01-13 cs.CV

Second-order Gaussian directional derivative representations for image high-resolution corner detection

Jiamiao Lu , Dongbo Xie , Junjie Qiu , Lingkun Ma , Changming Sun , Weichuan Zhang This is my paper
classification cs.CV
keywords corner detectionsecond-order Gaussian directional derivativeGaussian scale selectionadjacent cornershigh-resolution cornersEND-type corner modelL-type corner model
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that existing corner models fail for close-together corners because the grayscale profiles of adjacent corners interfere under Gaussian smoothing, so the filter response at a corner can be weaker than at the intervening edge. To fix this, it derives closed-form second-order Gaussian directional derivative (SOGDD) responses for two gray-value-based adjacent-corner models, the END-type and L-type, and compares the squared integral response at a corner with that at the midpoint edge. From that comparison it concludes that a corner is separable from an edge only when the Gaussian scale lies in the range σ∈(1,1.2), and it uses this to set σ²=1.2 in a new corner detector. The payoff would be a principled, parameter-light way to detect adjacent corners in high-resolution images, with better localization and matching performance than prior detectors.

Core claim

The authors establish that high-resolution (adjacent) corners can be accurately depicted only when the Gaussian filtering scale σ lies between 1 and 1.2, and they claim to be the first to derive this range from first principles. They do so by extending contour-based END-type and L-type corner models to gray-value-based models, deriving the SOGDD representations of a corner and of the edge point midway between two corners, and requiring the integrated squared response at the corner to exceed that at the edge. After scaling away the grayscale values, this condition becomes a quadratic inequality in σ whose roots, for the parameters shown, place the usable scale in (1,1.2); the limits come from

What carries the argument

The central object is the second-order Gaussian directional derivative (SOGDD) filter, the second derivative of a Gaussian taken along a rotated direction, convolved with the image to give a directional response Ψ(θ). The argument runs on an integral inequality: a corner is detectable when the integral of the squared SOGDD response over all filter orientations at the corner exceeds the same integral at the midpoint edge. For the two adjacent-corner intensity models, this inequality is reduced to a quadratic form in the scale σ, 8Aσ²−4√(2π)dCσ−πd²B>0, whose roots define the claimed scale range; the detector then uses an autocorrelation matrix built from multi-orientation SOGDD responses and a

Load-bearing premise

The claimed scale range σ∈(1,1.2) rests on the quadratic inequalities (13) and (15), but the coefficients A, B, and C in those inequalities are never defined or derived, so the transition from the exact integral inequality to the quoted scale range is an unshown algebraic step.

What would settle it

Numerically evaluate the integrals in Equations (5) and (7) for the paper's example values (d=3, α=π/8, β=π/3, T1=50, T2=100) and test whether inequality (12) holds for σ in (1,1.2) and fails for σ outside it, and whether the roots σ11=1.204 and σ13=1.253 match the actual sign change of the quadratic forms.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • A concrete, theoretically motivated scale (σ²=1.2) that lets a detector separate adjacent corners instead of merging them into an edge.
  • A demonstration that single-corner models are insufficient when corners are close, since adjacent corners' intensity profiles interfere under smoothing.
  • A high-resolution corner detector that improves localization error, repeatability under affine transforms, image matching, and 3D reconstruction over comparison methods.
  • The scale-selection condition is independent of grayscale values once scaled, meaning the corner/edge distinction is governed by geometry (d, α, β) rather than contrast.

Where Pith is reading between the lines

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

  • The derivation as presented leaves the coefficients A, B, C undefined; if they are recovered and the quadratic step is made explicit, the same inequality machinery could generate predicted scale ranges for other junction types (T, X, Y) and corner separations, not just the two models shown.
  • The narrow band (1,1.2) implies the method is effectively a fixed-scale detector; a testable extension is to check whether the optimal scale scales linearly with corner separation d, which the roots in (14) and (16) suggest, and whether multi-scale fusion could cover a wider range of d.
  • The paper's own criterion — corner response must exceed the midpoint edge response — could be turned into a diagnostic: on synthetic images of two corners with known d, α, β, detector performance should collapse sharply outside σ∈(1,1.2); observing that would confirm the mechanism, while observing good performance outside the range would falsify it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 8 minor

Summary. The paper studies second-order Gaussian directional derivative (SOGDD) representations for detecting high-resolution (adjacent) corners in images. It extends END-type and L-type corner models to gray-value-based models, derives SOGDD expressions (Eqs. (5)–(11)), and claims to establish for the first time the scale range σ∈(1,1.2) that enables accurate adjacent-corner detection. This range is derived from an integral inequality (Eq. (12)) that is reduced, without shown algebra, to quadratic inequalities whose coefficients are never defined. The proposed detector uses this scale range to fix σ²=1.2, computes a second-order directional derivative correlation matrix, and applies a product-over-trace corner metric. Experiments compare the method against recent detectors on ground-truth images, repeatability under affine/JPEG/noise transformations, HPatches matching, and 3D reconstruction. The paper concludes that the proposed method outperforms state-of-the-art in localization error, robustness, matching, and reconstruction.

Significance. If the theoretical derivation were complete and correct, the paper would offer a principled scale-selection rule for a genuinely difficult problem — detecting corners that are close enough to interfere with each other under Gaussian smoothing. The explicit SOGDD formulas for the two corner models and the targeted experimental protocol address a limitation of earlier corner models. However, the central theoretical claim (the scale range σ∈(1,1.2)) is currently unverifiable: the coefficients A1,B1,C1,A3,B3,C3 in Eqs. (13) and (15) are never defined, the reduction from Eq. (12) is not shown, and the reported numerical minima are presented as unexplained 'After verification' results. The paper also suffers an internal inconsistency between σ and σ² in the main parameter choice. Because these issues concern the core contribution rather than presentation, the manuscript in its present form does not provide a reliable basis for the proposed detector or its claimed advantages.

major comments (4)
  1. [§III.A, Eqs. (13)–(16)] The central result — the scale range σ∈(1,1.2) — is derived from the quadratic inequalities (13) and (15). Yet the coefficients A1,B1,C1,A3,B3,C3 are only described as functions of d, α, β; their explicit forms are never given. The transition from the integral inequality (12) to the quadratic forms (13)/(15) is not shown. The statements 'After verification' that yield minima 1.204 and 1.253 do not specify the parameter ranges or the algebra used. Without these steps, the derivation cannot be checked, so the paper's main theoretical contribution is unsupported rather than merely abridged.
  2. [§III.B and §IV] There is a load-bearing inconsistency in the Gaussian scale parameter. Section III.B states 'we deduce that the value of σ=1.2', while Section IV sets 'σ²=1.2'. The inequalities are quadratic in σ and the derived range is for σ. If σ=1.2, then σ²=1.44, which lies outside the claimed range; if σ²=1.2, then σ≈1.095, which is inside the range. The text does not use σ and σ² consistently, and this ambiguity directly affects the main parameter of the proposed detector and its connection to the derived range.
  3. [§III.A, Eq. (12)] The detectability criterion (12) is introduced with the reasoning that a corner is accurately detected only when its SOGDD is always greater than that at the edge point. The paper then replaces this pointwise condition with the inequality of integrals of squared SOGDDs. These are not equivalent, and no argument is given that the integral-of-squares criterion is necessary or sufficient for separating adjacent corners with the proposed detector (or any detector). The entire theoretical analysis rests on this unproven criterion; it should be justified or replaced by a condition that demonstrably matches the detector's decision rule.
  4. [§IV, Tables I–II and Fig. 8] The experimental evidence is mixed and lacks statistical validation. In Table I, on the 'Table' image the proposed method misses more corners than Harris (21 vs. 19) and produces more false corners than Zhang(2021) (21 vs. 9). In Table II, on Madrid Metropolis, Zhang(2023) reconstructs more sparse points (372K vs. 316K) and more dense points (1.35M vs. 1.31M) than the proposed method, despite the text claiming 'significant advantages'. The repeatability curves in Fig. 8 are shown without error bars or significance tests. The claimed overall superiority is therefore not uniformly supported by the presented numbers.
minor comments (8)
  1. [§II] 'Related et al. [60]' should be 'Revaud et al.'.
  2. [§IV.B] The shearing transform is described as 'uniform sampling in [1,1]' — likely a typo for [-1,1]. Also, 'with various scenes but no real scenes' is unclear; probably should be 'natural scenes' or similar.
  3. [§II and §III.A] The paper mentions 'three types of corner models' but only two models (END-type and L-type) are defined. Section III.A itself says 'two common high resolution corner models'.
  4. [Acknowledgments] The acknowledgments section contains placeholder text ('This should be a simple paragraph before the References to thank...'), indicating the manuscript is not finalized.
  5. [References] References [23] and [35] are the same work (Zhang and Sun, TPAMI 2021/2019); the duplicate should be removed.
  6. [§IV, Eq. (20)] The threshold Th=1×10^9 in the experiments is not explained. The metric Υ in Eq. (20) is a product-over-sum of eigenvalues; no analysis is given for how Th was chosen or how K (number of orientations) and the block size affect the results.
  7. [Fig. 8 caption] The caption says 'fifteen methods' while the text and figure list sixteen methods; please standardize.
  8. [§IV.C] For HPatches matching, the proposed method uses a 5-layer pyramid, SOSNet descriptor, and REKD direction filtering, while baseline results are taken from their papers with default parameters. This setup may not be directly comparable; the text should clarify or equalize the evaluation pipeline.

Circularity Check

0 steps flagged

No circular reduction found: the scale-range derivation is opaque (undefined A,B,C and an unshown quadratic step), but no equation is equivalent to its input by construction and the detector is benchmarked externally.

full rationale

The claimed derivation chain starts from the SOGDD filter (Eqs. 1-2), the convolution representation (Eq. 3), the END/L-type corner and edge models (Eqs. 4-11), and the detectability criterion (Eq. 12). The paper then asserts quadratic inequalities (13)/(15) with coefficients A1,B1,C1,A3,B3,C3 that are never defined, and reports numerical roots after 'verification' (1.204 and 1.253) that lead to the interval σ∈(1,1.2). This is a serious omitted derivation: without the coefficient formulas, the reader cannot check that Eq. (12) implies the stated range. However, an omitted algebraic step is not the same as a circular reduction. There is no place where a quantity is defined in terms of the very result it is supposed to predict, and no fitted parameter is renamed as a prediction. The lower endpoint σ>1 is imported from [23], a prior paper by the same authors, but it is a published, parameter-free noise-variance property of SOGDD filters, not a restatement of the adjacent-corner range, so under the review rules it counts as independent support rather than a load-bearing self-citation. Finally, the proposed detector is tested against external benchmarks (ground-truth localization, HPatches matching, repeatability under transformations, SfM/MVS reconstruction), so the method's empirical claims are not merely a renaming of the derivation's assumptions. The appropriate verdict is no significant circularity, with the caveat that the missing A,B,C derivation is a correctness/review problem rather than a circularity problem.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

The central claim (scale range for adjacent-corner detection) rests on a corner-model assumption, an unproven detectability criterion, and a quadratic inequality whose coefficients are never given. The lower bound comes from a self-cited theorem. No new physical entities are introduced.

free parameters (7)
  • Gaussian scale σ² = 1.2
    Chosen from the claimed range σ∈(1,1.2) in Section III.A; the range itself is derived from unstated coefficients and a specific example (d=3, α=π/8, β=π/3).
  • Distance d between adjacent corners = 3
    Set to 3 in the example to compute the quadratic roots (Section III.A); no evidence this generalizes.
  • Angles α and β = α=π/8, β=π/3
    Chosen for the example in Section III.A; the resulting σ roots depend on them.
  • Grayscale values T1, T2 = T1=50, T2=100
    Used to plot Figs 3-4; the paper claims invariance but does not show it.
  • Coefficients A, B, C = not given
    They appear in eqs (13)-(16) but are never defined; the full derivation of the inequalities is absent.
  • Number of orientations K = not given
    K appears in eq (17) but no value is specified in experiments.
  • Corner threshold Th = 1×10^9
    Set in Section IV; no sensitivity analysis.
axioms (4)
  • standard math Gaussian derivative filter model (eqs 1-2)
    Background definition of SOGDD used throughout.
  • domain assumption END-type and L-type corner models (eqs 4,6,8,10)
    The paper postulates these piecewise-constant gray-value models for adjacent corners; not justified from image statistics.
  • ad hoc to paper Integral inequality (12) as corner detectability criterion
    The paper asserts that a corner is detectable iff the sum of squared SOGDD integrals exceeds that of the edge point; no derivation or external support.
  • domain assumption Proof that σ²>1 reduces noise variance (from [23])
    Imported from prior self-cited work; used to set lower bound of the scale range.

pith-pipeline@v1.3.0-alltime-deepseek · 17540 in / 11148 out tokens · 91962 ms · 2026-08-03T10:50:49.385428+00:00 · methodology

0 comments
read the original abstract

Corner detection is widely used in various computer vision tasks, such as image matching and 3D reconstruction. Our research indicates that there are theoretical flaws in Zhang et al.'s use of a simple corner model to obtain a series of corner characteristics, as the grayscale information of two adjacent corners can affect each other. In order to address the above issues, a second-order Gaussian directional derivative (SOGDD) filter is used in this work to smooth two typical high-resolution angle models (i.e. END-type and L-type models). Then, the SOGDD representations of these two corner models were derived separately, and many characteristics of high-resolution corners were discovered, which enabled us to demonstrate how to select Gaussian filtering scales to obtain intensity variation information from images, accurately depicting adjacent corners. In addition, a new high-resolution corner detection method for images has been proposed for the first time, which can accurately detect adjacent corner points. The experimental results have verified that the proposed method outperforms state-of-the-art methods in terms of localization error, robustness to image blur transformation, image matching, and 3D reconstruction.

Figures

Figures reproduced from arXiv: 2601.08182 by Changming Sun, Dongbo Xie, Jiamiao Lu, Junjie Qiu, Lingkun Ma, Weichuan Zhang.

Figure 1
Figure 1. Figure 1: The results of image corner detection by the SOGDD detector [23] [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two common high resolution corner models: (a) END-type high [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The END-type high-resolution corner model with the SOGDD [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The L-type high-resolution corner model with the SOGDD represen [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 8
Figure 8. Figure 8: fig 8. It can be seen that among the following 16 methods [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 5
Figure 5. Figure 5: Test images (a) ‘Geometric’ and (b) ‘Table’ and their ground truths [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: The results of corner detection on the test image ‘Table’. (a) Harris, [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 6
Figure 6. Figure 6: The results of corner detection on the test image ‘Geometric’. (a) [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Average repeatability metrics for the fifteen methods under (a) [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison results of the mean matching accuracy for the seventeen [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. GloResNet: A lightweight 3D CNN with global topological features for preterm brain injury prediction

    cs.CV 2026-06 unverdicted novelty 4.0

    GloResNet, a ResNet-10-based lightweight 3D CNN pretrained on MedicalNet with global manifold mapping for topology preservation, achieves 75.18% average accuracy (peak 81.82%) in 5-fold cross-validation for preterm br...

Reference graph

Works this paper leans on

83 extracted references · 1 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Image feature information extraction for interest point detection: A comprehensive re- view,

    J. Jing, T. Gao, W. Zhang, Y . Gao, and C. Sun, “Image feature information extraction for interest point detection: A comprehensive re- view,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 4, pp. 4694–4712, 2022

    2022

  2. [2]

    Adaptive and generic corner detection based on the accelerated segment test,

    E. Mair, G. D. Hager, D. Burschka, M. Suppa, and G. Hirzinger, “Adaptive and generic corner detection based on the accelerated segment test,” inEuropean Conference on Computer Vision, 2010, pp. 183–196

    2010

  3. [3]

    Distinctive image features from scale-invariant keypoints,

    D. G. Lowe, “Distinctive image features from scale-invariant keypoints,” International Journal of Computer Vision, pp. 91–110, 2004

    2004

  4. [4]

    An iterative image registration technique with an application to stereo vision,

    B. D. Lucas and T. Kanade, “An iterative image registration technique with an application to stereo vision,” inInternational Joint Conference on Artificial Intelligence, 1981, pp. 674–679

    1981

  5. [5]

    Synthesis of interest point detectors through genetic programming,

    L. Trujillo and G. Olague, “Synthesis of interest point detectors through genetic programming,” inProceedings of the Conference on Genetic and Evolutionary Computation, 2006, pp. 887–894

    2006

  6. [6]

    Deep corner,

    S. Zhao, M. Gong, H. Zhao, J. Zhang, and D. Tao, “Deep corner,” International Journal of Computer Vision, pp. 2908–2932, 2023

    2023

  7. [7]

    CornerNet: Detecting objects as paired keypoints,

    H. Law and J. Deng, “CornerNet: Detecting objects as paired keypoints,” inProceedings of the European Conference on Computer Vision, 2018, pp. 734–750

    2018

  8. [8]

    Ecfrnet: Effective corner feature representations network for image corner detection,

    J. Jing, C. Liu, W. Zhang, Y . Gao, and C. Sun, “Ecfrnet: Effective corner feature representations network for image corner detection,”Expert Systems with Applications, vol. 211, p. 118673, 2023. Λ(nx, ny) =   p 2X i=− p 2 q 2X j=− q 2 ℓ2 1(nx +i, ny +j)· · · p 2X i=− p 2 q 2X j=− q 2 ℓ1(nx +i, ny +j)ℓ K (nx +i, ny +j) ... . . . ... p 2X i=− p ...

    2023

  9. [9]

    Corner detection based on shearlet transform and multi-directional structure tensor,

    M. Wang, W. Zhang, C. Sun, and A. Sowmya, “Corner detection based on shearlet transform and multi-directional structure tensor,”Pattern Recognition, vol. 103, p. 107299, 2020

    2020

  10. [10]

    A survey of corner detection methods,

    J. Wang and W. Zhang, “A survey of corner detection methods,” in2018 2nd International Conference on Electrical Engineering and Automation (ICEEA 2018). Atlantis Press, 2018, pp. 214–219

    2018

  11. [11]

    Multi-scale anisotropic gaussian kernels for image edge detection,

    Y . Li, Y . Bi, W. Zhang, and C. Sun, “Multi-scale anisotropic gaussian kernels for image edge detection,”IEEE Access, vol. 8, pp. 1803–1812, TABLE II COMPARISON RESULTS FORSFMANDMVS (†DENOTES THAT THE RESULT IS PROVIDED BY[83]). #Reg. #Sparse #Dense Track Reproj. Dataset Method Images Points Points Length Error (pixels) Fountain DoG † 11 16K 303K 4.910.4...

    2021

  12. [12]

    Aekan: Exploring superpixel-based autoencoder kolmogorov- arnold network for unsupervised multimodal change detection,

    T. Liu, J. Xu, T. Lei, Y . Wang, X. Du, W. Zhang, Z. Lv, and M. Gong, “Aekan: Exploring superpixel-based autoencoder kolmogorov- arnold network for unsupervised multimodal change detection,”IEEE Transactions on Geoscience and Remote Sensing, 2024

    2024

  13. [13]

    Few-shot fine-grained image classification: A comprehensive review,

    J. Ren, C. Li, Y . An, W. Zhang, and C. Sun, “Few-shot fine-grained image classification: A comprehensive review,”AI, vol. 5, no. 1, pp. 405–425, 2024

    2024

  14. [14]

    Ndpnet: A novel non-linear data projection network for few-shot fine-grained image classification,

    W. Zhang, X. Liu, Z. Xue, Y . Gao, and C. Sun, “Ndpnet: A novel non-linear data projection network for few-shot fine-grained image classification,”arXiv preprint arXiv:2106.06988, 2021

    Pith/arXiv arXiv 2021

  15. [15]

    Mutual interference mitigation of millimeter-wave radar based on variational mode decomposition and signal reconstruction,

    Y . Li, B. Feng, and W. Zhang, “Mutual interference mitigation of millimeter-wave radar based on variational mode decomposition and signal reconstruction,”Remote Sensing, vol. 15, no. 3, p. 557, 2023

    2023

  16. [16]

    Fast corner detection using approximate form of second-order gaussian directional derivative,

    T. Gao, J. Jing, C. Liu, W. Zhang, Y . Gao, and C. Sun, “Fast corner detection using approximate form of second-order gaussian directional derivative,”IEEE Access, vol. 8, pp. 194 092–194 104, 2020

    2020

  17. [17]

    A novel decision mechanism for image edge detection,

    J. Jing, S. Liu, C. Liu, T. Gao, W. Zhang, and C. Sun, “A novel decision mechanism for image edge detection,” inInternational Conference on Intelligent Computing. Springer, 2021, pp. 274–287

    2021

  18. [18]

    Image local structure information learning for fine-grained visual classification,

    J. Lu, W. Zhang, Y . Zhao, and C. Sun, “Image local structure information learning for fine-grained visual classification,”Scientific Reports, vol. 12, no. 1, p. 19205, 2022. 10

    2022

  19. [19]

    A corner detection method based on adaptive multi-directional anisotropic diffusion,

    J. Bao, J. Jing, W. Zhang, C. Liu, and T. Gao, “A corner detection method based on adaptive multi-directional anisotropic diffusion,”Multimedia Tools and Applications, vol. 81, no. 20, pp. 28 729–28 754, 2022

    2022

  20. [20]

    SSD: Single shot multibox detector,

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, and A. C. Berg, “SSD: Single shot multibox detector,” inEuropean Conference on Computer Vision, 2016, pp. 21–37

    2016

  21. [21]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” vol. 3, 2017

    2017

  22. [22]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European Conference on Computer Vision, 2020, pp. 213–229

    2020

  23. [23]

    Corner detection using second-order generalized Gaussian directional derivative representations,

    W. Zhang and C. Sun, “Corner detection using second-order generalized Gaussian directional derivative representations,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1213–1224, 2021

    2021

  24. [24]

    Corner detection using multi-directional structure tensor with multiple scales,

    ——, “Corner detection using multi-directional structure tensor with multiple scales,”International Journal of Computer Vision, pp. 438– 459, 2020

    2020

  25. [25]

    Image intensity variation information for interest point detection,

    W. Zhang, C. Sun, and Y . Gao, “Image intensity variation information for interest point detection,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 9883–9894, 2023

    2023

  26. [26]

    Discrete curvature representations for noise robust image corner detection,

    W. Zhang, C. Sun, T. Breckon, and N. Alshammari, “Discrete curvature representations for noise robust image corner detection,”IEEE Transac- tions on Image Processing, pp. 4444–4459, 2019

    2019

  27. [27]

    Recent advances on image edge detection: A comprehensive review,

    J. Jing, S. Liu, G. Wang, W. Zhang, and C. Sun, “Recent advances on image edge detection: A comprehensive review,”Neurocomputing, vol. 503, pp. 259–271, 2022

    2022

  28. [28]

    Pseudo-set frequency refinement architecture for fine-grained few-shot class-incremental learn- ing,

    Z. Pan, W. Zhang, X. Yu, M. Zhang, and Y . Gao, “Pseudo-set frequency refinement architecture for fine-grained few-shot class-incremental learn- ing,”Pattern Recognition, vol. 155, p. 110686, 2024

    2024

  29. [29]

    Re-abstraction and perturbing support pair network for few-shot fine-grained image classification,

    W. Zhang, Y . Zhao, Y . Gao, and C. Sun, “Re-abstraction and perturbing support pair network for few-shot fine-grained image classification,” Pattern Recognition, vol. 148, p. 110158, 2024

    2024

  30. [30]

    Corner detection and classification using anisotropic directional derivative representations,

    P.-L. Shui and W.-C. Zhang, “Corner detection and classification using anisotropic directional derivative representations,”IEEE Transactions on Image Processing, vol. 22, no. 8, pp. 3204–3218, 2013

    2013

  31. [31]

    Noise-robust edge detector combining isotropic and anisotropic gaussian kernels,

    ——, “Noise-robust edge detector combining isotropic and anisotropic gaussian kernels,”Pattern Recognition, vol. 45, no. 2, pp. 806–820, 2012

    2012

  32. [32]

    Noise robust image edge detection based upon the automatic anisotropic gaussian kernels,

    W. Zhang, Y . Zhao, T. P. Breckon, and L. Chen, “Noise robust image edge detection based upon the automatic anisotropic gaussian kernels,” Pattern Recognition, vol. 63, pp. 193–205, 2017

    2017

  33. [33]

    Contour-based corner detection via angle difference of principal directions of anisotropic gaussian directional derivatives,

    W.-C. Zhang and P.-L. Shui, “Contour-based corner detection via angle difference of principal directions of anisotropic gaussian directional derivatives,”Pattern Recognition, vol. 48, no. 9, pp. 2785–2797, 2015

    2015

  34. [34]

    Corner detection using gabor filters,

    W.-C. Zhang, F.-P. Wang, L. Zhu, and Z.-F. Zhou, “Corner detection using gabor filters,”IET Image Processing, vol. 8, no. 11, pp. 639–646, 2014

    2014

  35. [35]

    Corner detection using second-order generalized gaussian directional derivative representations,

    W. Zhang and C. Sun, “Corner detection using second-order generalized gaussian directional derivative representations,”IEEE transactions on pattern analysis and machine intelligence, vol. 43, no. 4, pp. 1213– 1224, 2019

    2019

  36. [36]

    Traffic flow digital twin generation for highway scenario based on radar-camera paired fusion,

    Y . Li and W. Zhang, “Traffic flow digital twin generation for highway scenario based on radar-camera paired fusion,”Scientific reports, vol. 13, no. 1, p. 642, 2023

    2023

  37. [37]

    Recurrent convolutional neural networks for 3d mandible segmentation in computed tomography,

    B. Qiu, J. Guo, J. Kraeima, H. H. Glas, W. Zhang, R. J. Borra, M. J. H. Witjes, and P. M. van Ooijen, “Recurrent convolutional neural networks for 3d mandible segmentation in computed tomography,”Journal of personalized medicine, vol. 11, no. 6, p. 492, 2021

    2021

  38. [38]

    An unbiased fea- ture estimation network for few-shot fine-grained image classification,

    J. Wang, J. Lu, J. Yang, M. Wang, and W. Zhang, “An unbiased fea- ture estimation network for few-shot fine-grained image classification,” Sensors, vol. 24, no. 23, p. 7737, 2024

    2024

  39. [39]

    Semi-supervised 3-d medical image segmentation using multiconsis- tency learning with fuzzy perception-guided target selection,

    T. Lei, W. Song, W. Zhang, X. Du, C. Li, L. He, and A. K. Nandi, “Semi-supervised 3-d medical image segmentation using multiconsis- tency learning with fuzzy perception-guided target selection,”IEEE Transactions on Radiation and Plasma Medical Sciences, vol. 9, no. 4, pp. 421–432, 2024

    2024

  40. [40]

    Asrsnet: Automatic salient region selection network for few-shot fine-grained image clas- sification,

    Y . Liao, W. Zhang, Y . Gao, C. Sun, and X. Yu, “Asrsnet: Automatic salient region selection network for few-shot fine-grained image clas- sification,” inInternational Conference on Pattern Recognition and Artificial Intelligence. Springer, 2022, pp. 627–638

    2022

  41. [41]

    Background-aware band selection for object tracking in hyperspectral videos,

    M. A. Islam, J. Zhou, W. Zhang, and Y . Gao, “Background-aware band selection for object tracking in hyperspectral videos,”IEEE Geoscience and Remote Sensing Letters, vol. 20, pp. 1–5, 2023

    2023

  42. [42]

    A principal component analysis-based feature optimization network for few-shot fine-grained image classification,

    M. Wang, B. Zheng, G. Wang, J. Yang, J. Lu, and W. Zhang, “A principal component analysis-based feature optimization network for few-shot fine-grained image classification,”Mathematics, vol. 13, no. 7, p. 1098, 2025

    2025

  43. [43]

    Fully unsupervised domain-agnostic image retrieval,

    Z. Zheng, H. Ren, Y . Wu, W. Zhang, H. Lu, Y . Yang, and H. T. Shen, “Fully unsupervised domain-agnostic image retrieval,”IEEE Transac- tions on Circuits and Systems for Video Technology, vol. 34, no. 6, pp. 5077–5090, 2023

    2023

  44. [44]

    Track-before-detect algorithm based on cost-reference particle filter bank for weak target detection,

    J. Lu, G. Peng, W. Zhang, and C. Sun, “Track-before-detect algorithm based on cost-reference particle filter bank for weak target detection,” IEEE Access, vol. 11, pp. 121 688–121 701, 2023

    2023

  45. [45]

    Dynamic accumulated attention map for interpreting evolution of decision-making in vision transformer,

    Y . Liao, Y . Gao, and W. Zhang, “Dynamic accumulated attention map for interpreting evolution of decision-making in vision transformer,”Pattern Recognition, vol. 165, p. 111607, 2025

    2025

  46. [46]

    Adaptive feature selection-based feature reconstruction network for few-shot learning,

    J. Ren, Y . An, T. Lei, J. Yang, W. Zhang, Z. Pan, Y . Liao, Y . Gao, C. Sun, and W. Zhang, “Adaptive feature selection-based feature reconstruction network for few-shot learning,”Pattern Recognition, 2025

    2025

  47. [47]

    Edge detection using multi-directional anisotropic gaussian directional derivative,

    Y . An, J. Jing, and W. Zhang, “Edge detection using multi-directional anisotropic gaussian directional derivative,”Signal, Image and Video Processing, vol. 17, no. 7, pp. 3767–3774, 2023

    2023

  48. [48]

    A combined corner and edge detector,

    C. Harris and M. Stephens, “A combined corner and edge detector,” in Alvey Vision Conference, 1988, pp. 10–5244

    1988

  49. [49]

    SUSAN: A new approach to low level image processing,

    S. M. Smith and J. M. Brady, “SUSAN: A new approach to low level image processing,”International Journal of Computer Vision, pp. 45–78, 1997

    1997

  50. [50]

    Faster and better: A machine learning approach to corner detection,

    E. Rosten, R. Porter, and T. Drummond, “Faster and better: A machine learning approach to corner detection,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 105–119, 2008

    2008

  51. [51]

    SURF: Speeded up robust features,

    H. Bay, “SURF: Speeded up robust features,” pp. 346–359, 2006

    2006

  52. [52]

    Anisotropic scale-invariant ellipse detection,

    Z. Wang, B. Zhong, and K.-K. Ma, “Anisotropic scale-invariant ellipse detection,”IEEE Transactions on Image Processing, 2024

    2024

  53. [53]

    Improving harris corner selection strategy,

    F. Bellavia, D. Tegolo, and C. Valenti, “Improving harris corner selection strategy,”IET Computer Vision, pp. 87–96, 2011

    2011

  54. [54]

    Harrisz +:harris corner selection for next- gen image matching pipelines,

    F. Bellavia and D. Mishkin, “Harrisz +:harris corner selection for next- gen image matching pipelines,”Pattern Recognition Letters, pp. 141– 147, 2022

    2022

  55. [55]

    LF-Net: Learning local features from images,

    Y . Ono, E. Trulls, P. Fua, and K. M. Yi, “LF-Net: Learning local features from images,” 2018

    2018

  56. [56]

    Superpoint: Self- supervised interest point detection and description,

    D. DeTone, T. Malisiewicz, and A. Rabinovich, “Superpoint: Self- supervised interest point detection and description,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2018, pp. 224–236

    2018

  57. [57]

    D2-Net: A trainable CNN for joint description and detection of local features,

    M. Dusmanu, I. Rocco, T. Pajdla, M. Pollefeys, J. Sivic, A. Torii, and T. Sattler, “D2-Net: A trainable CNN for joint description and detection of local features,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019, pp. 8092–8101

    2019

  58. [58]

    Benchmarking 6DOF outdoor visual localization in changing conditions,

    T. Sattler, W. Maddern, C. Toft, A. Torii, L. Hammarstrand, E. Stenborg, D. Safari, M. Okutomi, M. Pollefeys, J. Sivicet al., “Benchmarking 6DOF outdoor visual localization in changing conditions,” inPro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 8601–8610

    2018

  59. [59]

    InLoc: Indoor visual localization with dense matching and view synthesis,

    H. Taira, M. Okutomi, T. Sattler, M. Cimpoi, M. Pollefeys, J. Sivic, T. Pajdla, and A. Torii, “InLoc: Indoor visual localization with dense matching and view synthesis,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 7199–7209

    2018

  60. [60]

    R2D2: Reliable and repeatable detector and descriptor,

    J. Revaud, C. De Souza, M. Humenberger, and P. Weinzaepfel, “R2D2: Reliable and repeatable detector and descriptor,” 2019

    2019

  61. [61]

    ZippyPoint: Fast interest point detection, description, and matching through mixed precision discretization,

    M. Kanakis, S. Maurer, M. Spallanzani, A. Chhatkuli, and L. Van Gool, “ZippyPoint: Fast interest point detection, description, and matching through mixed precision discretization,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2023, pp. 6114–6123

    2023

  62. [62]

    DeDoDe: Detect, don’t describe-describe, don’t detect for local feature matching,

    J. Edstedt, G. B ¨okman, M. Wadenb ¨ack, and M. Felsberg, “DeDoDe: Detect, don’t describe-describe, don’t detect for local feature matching,” inInternational Conference on 3D Vision, 2024, pp. 148–157

    2024

  63. [63]

    BALF: Simple and efficient blur aware local feature detector,

    Z. Zhao, “BALF: Simple and efficient blur aware local feature detector,” inProceedings of the IEEE Winter Conference on Applications of Computer Vision, 2024, pp. 3362–3372

    2024

  64. [64]

    Dynamic keypoint detection network for image matching,

    Y . Gao, J. He, T. Zhang, Z. Zhang, and Y . Zhang, “Dynamic keypoint detection network for image matching,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  65. [65]

    Xfeat: Accelerated features for lightweight image matching,

    G. Potje, F. Cadar, A. Araujo, R. Martins, and E. R. Nascimento, “Xfeat: Accelerated features for lightweight image matching,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2024, pp. 2682–2691

    2024

  66. [66]

    Alike: Accurate and lightweight keypoint detection and descriptor extraction,

    X. Zhao, X. Wu, J. Miao, W. Chen, P. C. Chen, and Z. Li, “Alike: Accurate and lightweight keypoint detection and descriptor extraction,” IEEE Transactions on Multimedia, pp. 3101–3112, 2022

    2022

  67. [67]

    Aliked: A lighter keypoint and descriptor extraction network via deformable trans- 11 formation,

    X. Zhao, X. Wu, W. Chen, P. C. Chen, Q. Xu, and Z. Li, “Aliked: A lighter keypoint and descriptor extraction network via deformable trans- 11 formation,”IEEE Transactions on Instrumentation and Measurement, pp. 1–16, 2023

    2023

  68. [68]

    Robust image corner detection based on the chord-to-point distance accumulation technique,

    M. Awrangjeb and G. Lu, “Robust image corner detection based on the chord-to-point distance accumulation technique,”IEEE Transactions on Multimedia, pp. 1059–1072, 2008

    2008

  69. [69]

    Accurate junction detection and characterization in natural images,

    G.-S. Xia, J. Delon, and Y . Gousseau, “Accurate junction detection and characterization in natural images,”International Journal of Computer Vision, pp. 31–56, 2014

    2014

  70. [70]

    Scale & affine invariant interest point detectors,

    K. Mikolajczyk and C. Schmid, “Scale & affine invariant interest point detectors,”International Journal of Computer Vision, pp. 63–86, 2004

    2004

  71. [71]

    Corner detection and classification using anisotropic directional derivative representations,

    P.-L. Shui and W.-C. Zhang, “Corner detection and classification using anisotropic directional derivative representations,”IEEE Transactions on Image Processing, pp. 3204–3218, 2013

    2013

  72. [72]

    LIFT: Learned invariant feature transform,

    K. M. Yi, E. Trulls, V . Lepetit, and P. Fua, “LIFT: Learned invariant feature transform,” inEuropean Conference on Computer Vision, 2016, pp. 467–483

    2016

  73. [73]

    Learning discrimina- tive and transformation covariant local feature detectors,

    X. Zhang, F. X. Yu, S. Karaman, and S.-F. Chang, “Learning discrimina- tive and transformation covariant local feature detectors,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 6818–6826

    2017

  74. [74]

    KAZE features,

    P. F. Alcantarilla, A. Bartoli, and A. J. Davison, “KAZE features,” in European Conference on Computer Vision, 2012, pp. 214–227

    2012

  75. [75]

    Key.Net: Keypoint detection by handcrafted and learned CNN filters revisited,

    A. Barroso-Laguna and K. Mikolajczyk, “Key.Net: Keypoint detection by handcrafted and learned CNN filters revisited,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 698–711, 2023

    2023

  76. [76]

    Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors,

    V . Balntas, K. Lenc, A. Vedaldi, and K. Mikolajczyk, “Hpatches: A benchmark and evaluation of handcrafted and learned local descriptors,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 5173–5182

    2017

  77. [77]

    ASLFeat: Learning local features of accurate shape and localization,

    Z. Luo, L. Zhou, X. Bai, H. Chen, J. Zhang, Y . Yao, S. Li, T. Fang, and L. Quan, “ASLFeat: Learning local features of accurate shape and localization,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2020, pp. 6589–6598

    2020

  78. [78]

    Self-supervised equivariant learning for oriented keypoint detection,

    J. Lee, B. Kim, and M. Cho, “Self-supervised equivariant learning for oriented keypoint detection,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2022, pp. 4847–4857

    2022

  79. [79]

    Com- parative evaluation of hand-crafted and learned local features,

    J. L. Schonberger, H. Hardmeier, T. Sattler, and M. Pollefeys, “Com- parative evaluation of hand-crafted and learned local features,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 1482–1491

    2017

  80. [80]

    Structure-from-motion revisited,

    J. L. Schonberger and J.-M. Frahm, “Structure-from-motion revisited,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 4104–4113

    2016

Showing first 80 references.