The reviewed record of science sign in
Pith

arxiv: 2606.00715 · v1 · pith:RKUSGVXB · submitted 2026-05-30 · stat.ME · stat.ML

Rate-optimal neural boundary detection from unlabeled noisy images

Kyeongho Kim , Ilsang Ohn This is my paper

Reviewed by Pith2026-06-28 18:27 UTCgrok-4.3pith:RKUSGVXBopen to challenge →

classification stat.ME stat.ML
keywords boundary detectionunsupervised learningdeep neural networkssurrogate lossminimax ratepiecewise smooth boundariesnoisy images
0
0 comments X

The pith

A deep neural network estimator with a continuous hinge surrogate loss recovers object boundaries from noisy unlabeled images at the minimax optimal rate under piecewise smooth models.

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

The paper develops a method for recovering object regions from raw noisy image intensities without any pixel labels or parametric intensity models. It introduces a continuous hinge-type surrogate loss that is Fisher consistent under a mild separation assumption on the intensity distributions and yields a calibration inequality to the symmetric difference error. The loss combines with deep neural networks to represent complex boundaries. Under a piecewise smooth boundary model that allows corners and kinks, the resulting estimator attains the minimax-optimal recovery rate up to logarithmic factors. Numerical experiments confirm stable performance across noise levels and shapes.

Core claim

Under a piecewise smooth boundary model, the deep neural network estimator based on the proposed continuous hinge surrogate loss achieves the minimax-optimal boundary recovery rate, up to logarithmic factors. The method recovers an unknown object region from raw intensity observations without labels or parametric intensity models, and the piecewise smooth formulation accommodates boundaries with corners and kinks.

What carries the argument

Continuous hinge-type surrogate loss for boundary detection, which is Fisher consistent under mild separation of intensity distributions and amenable to gradient-based optimization with deep neural networks.

If this is right

  • The estimator extends beyond globally smooth boundary models by handling corners and kinks.
  • The calibration inequality directly links excess surrogate risk to the symmetric difference error of the recovered region.
  • The approach works without pixel-wise annotations or parametric intensity models.
  • Numerical results show accurate recovery across noise levels and shape configurations.

Where Pith is reading between the lines

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

  • The method could apply to domains such as medical imaging where labels are costly to obtain.
  • Extensions might test the loss on higher-dimensional or temporal image data.
  • The rate-optimality result suggests comparisons to parametric boundary models on the same data.

Load-bearing premise

The mild separation assumption on the intensity distributions required for Fisher consistency of the surrogate loss.

What would settle it

An experiment or dataset where the intensity distributions violate the separation assumption and the estimated region fails to converge to the true boundary at the claimed rate.

Figures

Figures reproduced from arXiv: 2606.00715 by Ilsang Ohn, Kyeongho Kim.

Figure 1
Figure 1. Figure 1: Qualitative comparison of boundary detection results in representative synthetic scenarios. [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Symmetric difference error (SDE) under Gaussian noise for the synthetic examples. The error [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Symmetric difference error (SDE) under Poisson noise for the synthetic examples. The error [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison on cropped BBBC038 microscopy images under different levels of [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Symmetric difference error on cropped BBBC038 microscopy images under different Gaussian [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
read the original abstract

We study boundary detection for unlabeled noisy images from a statistical perspective. The aim is to recover an unknown object region from raw intensity observations without pixel-wise annotating labels or a parametric model for the intensity distributions. Motivated by robust Gibbs posterior approaches based on thresholded misclassification losses, we propose a continuous hinge-type surrogate loss for boundary detection. The proposed loss is amenable to gradient-based optimization and can be combined with deep neural networks to represent complex object boundaries. We prove that the proposed loss function is Fisher consistent under a mild separation assumption and obtain a calibration inequality linking excess surrogate risk to the symmetric difference error of the estimated region. Under a piecewise smooth boundary model, we prove that the resulting deep neural network estimator achieves the minimax-optimal boundary recovery rate, up to logarithmic factors. The piecewise smooth formulation accommodates boundaries with corners and kinks, thereby extending beyond globally smooth boundary models. Numerical experiments demonstrate that the proposed method accurately and stably recovers object boundaries across a range of noise levels and shape configurations, and compares favorably with existing unsupervised boundary detection methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a continuous hinge-type surrogate loss for unsupervised boundary detection from unlabeled noisy images. It claims Fisher consistency of the loss under a mild separation assumption on the conditional intensity distributions, derives a calibration inequality relating excess surrogate risk to symmetric-difference error of the estimated region, and proves that a deep neural network estimator attains the minimax-optimal boundary recovery rate (up to logarithmic factors) under a piecewise-smooth boundary model that permits corners and kinks. Numerical experiments are presented to illustrate performance across noise levels and shapes.

Significance. If the derivations hold, the work supplies a statistically grounded, rate-optimal DNN method for boundary recovery without labels or parametric intensity models, extending beyond globally smooth boundaries. The combination of surrogate-loss calibration with minimax analysis under a flexible boundary class is a clear strength, as is the explicit handling of the piecewise-smooth setting.

major comments (2)
  1. [Abstract / theoretical results section] Abstract and theoretical development: the Fisher-consistency claim, calibration inequality, and subsequent minimax rate all rest on the mild separation assumption on intensity distributions. The piecewise-smooth boundary model alone does not guarantee this separation; if the conditional distributions inside and outside the object overlap strongly, consistency fails and the rate result does not follow. The manuscript should state the precise form of the separation condition (e.g., a quantitative lower bound on the difference of conditional expectations or densities) and discuss whether it can be checked from data or relaxed.
  2. [Theoretical results section] Theoretical results section: the minimax optimality statement is obtained only after invoking external lower bounds and the separation assumption. Without an explicit statement of how the DNN approximation error, the surrogate excess risk, and the separation condition combine to yield the final rate (including the precise logarithmic factors), the optimality claim cannot be verified from the given material.
minor comments (2)
  1. [Numerical experiments] The numerical experiments section would benefit from explicit reporting of the precise noise variances, boundary smoothness parameters, and number of Monte Carlo replications used to generate the reported comparisons.
  2. [Methods] Notation for the symmetric-difference error and the surrogate loss should be introduced with a single consistent symbol set in the methods section rather than redefined inline.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address the two major comments below and will revise the manuscript to improve clarity on the separation assumption and the rate derivation.

read point-by-point responses

  1. Referee: [Abstract / theoretical results section] Abstract and theoretical development: the Fisher-consistency claim, calibration inequality, and subsequent minimax rate all rest on the mild separation assumption on intensity distributions. The piecewise-smooth boundary model alone does not guarantee this separation; if the conditional distributions inside and outside the object overlap strongly, consistency fails and the rate result does not follow. The manuscript should state the precise form of the separation condition (e.g., a quantitative lower bound on the difference of conditional expectations or densities) and discuss whether it can be checked from data or relaxed.

    Authors: We agree that the separation condition requires a more explicit quantitative statement. In the revised manuscript we will define it precisely as a positive lower bound δ on the difference of conditional expectations |μ_in(x) − μ_out(x)| ≥ δ for x near the boundary (with an analogous density version if needed), and we will add a short paragraph discussing its role, the fact that it is standard for Fisher consistency of surrogate losses, and the practical difficulty of verifying it from unlabeled data. We note that the assumption is already implicit in the calibration inequality derived in the paper; relaxing it would require a different loss or analysis that is outside the current scope. revision: yes

  2. Referee: [Theoretical results section] Theoretical results section: the minimax optimality statement is obtained only after invoking external lower bounds and the separation assumption. Without an explicit statement of how the DNN approximation error, the surrogate excess risk, and the separation condition combine to yield the final rate (including the precise logarithmic factors), the optimality claim cannot be verified from the given material.

    Authors: We accept that the chaining of error terms should be written out more explicitly. The revision will insert a dedicated paragraph (or short subsection) that (i) recalls the DNN approximation rate for the piecewise-smooth boundary class, (ii) invokes the calibration inequality whose constant depends on the separation parameter δ, (iii) absorbs the resulting excess symmetric-difference risk into the final bound, and (iv) combines it with the external minimax lower bound, making the precise logarithmic factors (arising from the DNN entropy integral and the usual log n terms in the rate) transparent. This will allow direct verification without external reconstruction. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on external minimax bounds and explicit assumptions

full rationale

The derivation proceeds by stating a mild separation assumption, proving Fisher consistency of the surrogate loss and a calibration inequality to symmetric-difference error, then establishing the DNN estimator's rate under the piecewise-smooth boundary model. These steps are presented as theorems relying on the stated assumption and external minimax lower bounds rather than any self-referential definition, fitted parameter renamed as prediction, or self-citation chain that reduces the central claim to its own inputs by construction. The paper is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The optimality claim rests on the piecewise-smooth boundary model and the mild separation assumption; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption Mild separation assumption on intensity distributions
    Invoked to obtain Fisher consistency of the proposed surrogate loss.
  • domain assumption Piecewise smooth boundary model
    Used to establish the minimax-optimal rate for the DNN estimator.

pith-pipeline@v0.9.1-grok · 5707 in / 1195 out tokens · 20401 ms · 2026-06-28T18:27:12.891866+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

93 extracted references · 3 canonical work pages

  1. [1]

    The Annals of Statistics , volume=

    Robust and rate-optimal Gibbs posterior inference on the boundary of a noisy image , author=. The Annals of Statistics , volume=. 2020 , publisher=

    2020

  2. [2]

    The Annals of Statistics , volume=

    Bayesian detection of image boundaries , author=. The Annals of Statistics , volume=

  3. [3]

    Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume=

    Local likelihood tracking of fault lines and boundaries , author=. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , volume=. 2001 , publisher=

    2001

  4. [4]

    The Annals of Statistics , pages=

    Asymptotical minimax recovery of sets with smooth boundaries , author=. The Annals of Statistics , pages=. 1995 , publisher=

    1995

  5. [5]

    Adaptive variational

    Ohn, Ilsang and Lin, Lizhen , journal=. Adaptive variational. 2024 , publisher=

    2024

  6. [6]

    Electronic Journal of Statistics , volume=

    Intrinsic and extrinsic deep learning on manifolds , author=. Electronic Journal of Statistics , volume=. 2024 , publisher=

    2024

  7. [7]

    Journal of the American Statistical Association , volume=

    Factor augmented sparse throughput deep relu neural networks for high dimensional regression , author=. Journal of the American Statistical Association , volume=. 2024 , publisher=

    2024

  8. [8]

    Electronic Journal of Statistics , volume=

    Convergence rates of deep ReLU networks for multiclass classification , author=. Electronic Journal of Statistics , volume=. 2022 , publisher=

    2022

  9. [9]

    arXiv preprint arXiv:2506.14899 , year=

    Optimal Convergence Rates of Deep Neural Network Classifiers , author=. arXiv preprint arXiv:2506.14899 , year=

    work page arXiv

  10. [10]

    The Annals of Statistics , volume=

    Statistical behavior and consistency of classification methods based on convex risk minimization , author=. The Annals of Statistics , volume=. 2004 , publisher=

    2004

  11. [11]

    Journal of the American Statistical Association , volume=

    Convexity, classification, and risk bounds , author=. Journal of the American Statistical Association , volume=. 2006 , publisher=

    2006

  12. [12]

    The annals of statistics , volume=

    Additive logistic regression: a statistical view of boosting (with discussion and a rejoinder by the authors) , author=. The annals of statistics , volume=. 2000 , publisher=

    2000

  13. [13]

    The Annals of Statistics , volume=

    Statistical performance of support vector machines , author=. The Annals of Statistics , volume=. 2008 , publisher=

    2008

  14. [14]

    The Annals of Statistics , volume=

    Fast rates for support vector machines using Gaussian kernels , author=. The Annals of Statistics , volume=. 2007 , publisher=

    2007

  15. [15]

    Bernoulli , volume=

    Classifiers of support vector machine type with ell1 complexity regularization , author=. Bernoulli , volume=. 2006 , publisher=

    2006

  16. [16]

    2008 , publisher=

    Support vector machines , author=. 2008 , publisher=

    2008

  17. [17]

    Statistics & probability letters , volume=

    A note on margin-based loss functions in classification , author=. Statistics & probability letters , volume=. 2004 , publisher=

    2004

  18. [18]

    Journal of Machine Learning Research , volume=

    Classification with deep neural networks and logistic loss , author=. Journal of Machine Learning Research , volume=

  19. [19]

    International Conference on Machine Learning , pages=

    Cross-entropy loss functions: Theoretical analysis and applications , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  20. [20]

    International conference on machine learning , pages=

    Masked Bayesian neural networks: Theoretical guarantee and its posterior inference , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  21. [21]

    Bernoulli , volume=

    Adaptive deep learning for nonlinear time series models , author=. Bernoulli , volume=. 2025 , publisher=

    2025

  22. [22]

    Electronic Journal of Statistics , volume=

    Oracle posterior contraction rates under hierarchical priors , author=. Electronic Journal of Statistics , volume=. 2021 , publisher=

    2021

  23. [23]

    The Annals of Statistics , volume=

    Convergence rates of variational posterior distributions , author=. The Annals of Statistics , volume=

  24. [24]

    The Annals of Statistics , volume=

    Concentration of tempered posteriors and of their variational approximations , author=. The Annals of Statistics , volume=. 2020 , publisher=

    2020

  25. [25]

    The Annals of Statistics , volume=

    -variational inference with statistical guarantees , author=. The Annals of Statistics , volume=. 2020 , publisher=

    2020

  26. [26]

    On statistical optimality of variational

    Pati, Debdeep and Bhattacharya, Anirban and Yang, Yun , booktitle=. On statistical optimality of variational. 2018 , organization=

    2018

  27. [27]

    Consistency of

    Ch. Consistency of. Symposium on Advances in Approximate Bayesian Inference , pages=. 2019 , organization=

    2019

  28. [28]

    A general framework for

    Gao, Chao and van der Vaart, Aad and Zhou, Harrison H , journal=. A general framework for. 2020 , publisher=

    2020

  29. [29]

    Consistent

    Jiang, Sheng and Tokdar, Surya T , journal=. Consistent

  30. [30]

    The Annals of Statistics , volume=

    Bayesian fractional posteriors , author=. The Annals of Statistics , volume=. 2019 , publisher=

    2019

  31. [31]

    On universal

    Lember, J. On universal. Statistics and Decisions-International Journal Stochastic Methods and Models , volume=. 2007 , publisher=

    2007

  32. [32]

    Nonparametric

    Ghosal, Subhashis and Lember, J. Nonparametric. Electronic Journal of Statistics , volume=. 2008 , publisher=

    2008

  33. [33]

    Adaptive

    Belitser, Eduard and Ghosal, Subhashis , journal=. Adaptive. 2003 , publisher=

    2003

  34. [34]

    Fundamentals of nonparametric

    Ghosal, Subhashis and Van der Vaart, Aad , volume=. Fundamentals of nonparametric. 2017 , publisher=

    2017

  35. [35]

    Advances in statistical decision theory and applications , pages=

    Non-informative priors via sieves and packing numbers , author=. Advances in statistical decision theory and applications , pages=. 1997 , publisher=

    1997

  36. [36]

    The Annals of Statistics , volume=

    Convergence rates of posterior distributions , author=. The Annals of Statistics , volume=. 2000 , publisher=

    2000

  37. [37]

    The Annals of Statistics , volume=

    Convergence rates of posterior distributions for noniid observations , author=. The Annals of Statistics , volume=. 2007 , publisher=

    2007

  38. [38]

    Rate exact

    Gao, Chao and Zhou, Harrison H , journal=. Rate exact. 2016 , publisher=

    2016

  39. [39]

    The Annals of Statistics , volume=

    On adaptive posterior concentration rates , author=. The Annals of Statistics , volume=. 2015 , publisher=

    2015

  40. [40]

    Rates of convergence for sparse variational

    Burt, David and Rasmussen, Carl Edward and Van Der Wilk, Mark , booktitle=. Rates of convergence for sparse variational. 2019 , organization=

    2019

  41. [41]

    Ohn, Ilsang and Lin, Lizhen , journal=. Optimal

  42. [42]

    Bayesian Analysis , volume=

    Posterior consistency of factor dimensionality in high-dimensional sparse factor models , author=. Bayesian Analysis , volume=. 2022 , publisher=

    2022

  43. [43]

    Statistical learning theory and stochastic optimization: Ecole d'Et

    Catoni, Olivier , volume=. Statistical learning theory and stochastic optimization: Ecole d'Et. 2004 , publisher=

    2004

  44. [44]

    IEEE Transactions on Information Theory , volume=

    Information-theoretic upper and lower bounds for statistical estimation , author=. IEEE Transactions on Information Theory , volume=. 2006 , publisher=

    2006

  45. [45]

    From -entropy to

    Zhang, Tong , journal=. From -entropy to. 2006 , publisher=

    2006

  46. [46]

    The Journal of Machine Learning Research , volume=

    On the properties of variational approximations of Gibbs posteriors , author=. The Journal of Machine Learning Research , volume=. 2016 , publisher=

    2016

  47. [47]

    User-friendly introduction to

    Alquier, Pierre , journal=. User-friendly introduction to

  48. [48]

    The Annals of Statistics , volume=

    On the contraction properties of some high-dimensional quasi-posterior distributions , author=. The Annals of Statistics , volume=. 2017 , publisher=

    2017

  49. [49]

    An approach to large-scale quasi-

    Atchad. An approach to large-scale quasi-. arXiv preprint arXiv:1803.10282 , year=

    work page arXiv

  50. [50]

    The Annals of Statistics , volume=

    Needles and straw in a haystack: Posterior concentration for possibly sparse sequences , author=. The Annals of Statistics , volume=. 2012 , publisher=

    2012

  51. [51]

    The Annals of Statistics , volume=

    Bayesian linear regression with sparse priors , author=. The Annals of Statistics , volume=. 2015 , publisher=

    2015

  52. [52]

    Empirical

    Martin, Ryan and Mess, Raymond and Walker, Stephen G , journal=. Empirical. 2017 , publisher=

    2017

  53. [53]

    The Annals of Statistics , volume=

    Adaptive learning rates for support vector machines working on data with low intrinsic dimension , author=. The Annals of Statistics , volume=. 2021 , publisher=

    2021

  54. [54]

    Nonparametric regression using deep neural networks with

    Schmidt-Hieber, Johannes , journal=. Nonparametric regression using deep neural networks with. 2020 , publisher=

    2020

  55. [55]

    Journal of Machine Learning Research , volume=

    Advantage of Deep Neural Networks for Estimating Functions with Singularity on Hypersurfaces , author=. Journal of Machine Learning Research , volume=

  56. [56]

    The Annals of Statistics , volume=

    Deep nonparametric regression on approximate manifolds: Nonasymptotic error bounds with polynomial prefactors , author=. The Annals of Statistics , volume=. 2023 , publisher=

    2023

  57. [57]

    Information and Inference: A Journal of the IMA , volume=

    Nonparametric regression on low-dimensional manifolds using deep ReLU networks: Function approximation and statistical recovery , author=. Information and Inference: A Journal of the IMA , volume=. 2022 , publisher=

    2022

  58. [58]

    Neural Networks , volume=

    Fast convergence rates of deep neural networks for classification , author=. Neural Networks , volume=. 2021 , publisher=

    2021

  59. [59]

    Neural Computation , volume=

    Nonconvex sparse regularization for deep neural networks and its optimality , author=. Neural Computation , volume=. 2022 , publisher=

    2022

  60. [60]

    The Annals of Statistics , volume=

    On the rate of convergence of fully connected deep neural network regression estimates , author=. The Annals of Statistics , volume=. 2021 , publisher=

    2021

  61. [61]

    SIAM Journal on Mathematical Analysis , volume=

    Deep network approximation for smooth functions , author=. SIAM Journal on Mathematical Analysis , volume=. 2021 , publisher=

    2021

  62. [62]

    Communications in Computational Physics , volume=

    Deep network approximation characterized by number of neurons , author=. Communications in Computational Physics , volume=

  63. [63]

    Neural Networks , volume=

    Nonlinear approximation via compositions , author=. Neural Networks , volume=. 2019 , publisher=

    2019

  64. [64]

    International Conference on Learning Representations , year=

    Adam: A method for stochastic optimization , author=. International Conference on Learning Representations , year=

  65. [65]

    Proceedings of the 32nd International Conference on Neural Information Processing Systems , pages=

    Posterior concentration for sparse deep learning , author=. Proceedings of the 32nd International Conference on Neural Information Processing Systems , pages=

  66. [66]

    Proceedings of the 37th International Conference on Machine Learning , pages =

    Convergence Rates of Variational Inference in Sparse Deep Learning , author =. Proceedings of the 37th International Conference on Machine Learning , pages =. 2020 , publisher =

    2020

  67. [67]

    arXiv preprint arXiv:2206.00853 , year=

    Masked Bayesian Neural Networks: Computation and Optimality , author=. arXiv preprint arXiv:2206.00853 , year=

    work page arXiv

  68. [68]

    On the expressiveness of approximate inference in

    Foong, Andrew and Burt, David and Li, Yingzhen and Turner, Richard , booktitle =. On the expressiveness of approximate inference in

  69. [69]

    The Annals of Statistics , volume=

    Minimax-optimal nonparametric regression in high dimensions , author=. The Annals of Statistics , volume=. 2015 , publisher=

    2015

  70. [70]

    The Annals of Statistics , volume=

    Linear Smoothers and Additive Models , author=. The Annals of Statistics , volume=

  71. [71]

    The Annals of Statistics , volume=

    Optimal Global Rates of Convergence for Nonparametric Regression , author=. The Annals of Statistics , volume=. 1982 , publisher=

    1982

  72. [72]

    2012 , publisher=

    Statistical inference for spatial Poisson processes , author=. 2012 , publisher=

    2012

  73. [73]

    The Annals of Statistics , volume=

    On asymptotically optimal confidence regions and tests for high-dimensional models , author=. The Annals of Statistics , volume=. 2014 , publisher=

    2014

  74. [74]

    Adaptive

    Shen, Weining and Tokdar, Surya T and Ghosal, Subhashis , journal=. Adaptive. 2013 , publisher=

    2013

  75. [75]

    Variational

    Ray, Kolyan and Szab. Variational. Journal of the American Statistical Association , volume=. 2022 , publisher=

    2022

  76. [76]

    Proceedings of the 34th International Conference on Neural Information Processing Systems , volume=

    Efficient variational inference for sparse deep learning with theoretical guarantee , author=. Proceedings of the 34th International Conference on Neural Information Processing Systems , volume=

  77. [77]

    Journal of the American Statistical Association , volume=

    Variable selection via Gibbs sampling , author=. Journal of the American Statistical Association , volume=. 1993 , publisher=

    1993

  78. [78]

    Journal of the American Statistical Association , volume=

    Adaptive conditional distribution estimation with Bayesian decision tree ensembles , author=. Journal of the American Statistical Association , volume=. 2023 , publisher=

    2023

  79. [79]

    2009 , publisher=

    Optimal transport: old and new , author=. 2009 , publisher=

    2009

  80. [80]

    Stanford Artificial Intelligence Project , year=

    A 3×3 isotropic gradient operator for image processing , author=. Stanford Artificial Intelligence Project , year=

Showing first 80 references.