Pith. sign in

REVIEW 4 major objections 5 minor 44 references

Computational Microstructure Analysis of Sintered Ceramics

T0 review · 4 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read An automated pipeline segments sintered ceramic SEM images and extracts microstructure statistics in agreement with expert manual analysis.

desk verdict The pipeline is a reasonable engineering effort, but its validation is impossible: the reported IoU at 1400°C cannot coexist with the reported porosity values. read the letter →

arxiv 2608.04216 v1 pith:XQK3OJZO submitted 2026-08-04 cond-mat.mtrl-sci cs.CE

Nikhil Dhanankam , Deeksha Kodangal , Rajendra K. Bordia , Ulf D. Schiller This is my paper · ORCID
classification cond-mat.mtrl-scics.CE
keywords imageanalysismicrostructurecharacterizationadditivemanufacturingscanningelectronmicroscopythresholdingporositygrainsizesinteredceramics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper aims to automate the slow, subjective work of measuring porosity, solid fraction, grain size, and pore size from scanning electron microscopy images of sintered ceramics. The obstacle is that grain and pore phases overlap in grayscale intensity, producing unimodal histograms that global thresholding cannot separate. The proposed workflow combines PerSplat topological filtering with Sauvola local adaptive thresholding and morphological reconstruction, then validates the results against manual ImageJ segmentations. The authors report Intersection over Union scores of 95.14% and 99.85% in the abstract, with mean values of 92.18% and 99.35% in the results table, and physical property estimates that track the manual reference data. If the agreement holds, the workflow gives materials science a high-throughput path to the annotated microstructure datasets needed for machine learning.

What carries the argument

The load-bearing combination is PerSplat topological filtering plus Sauvola local adaptive thresholding, followed by morphological reconstruction for pore identification. PerSplat uses persistent homology to discard noise while keeping topologically significant features; Sauvola sets a threshold from the local mean and standard deviation in a window around each pixel, which is what makes binarization work for unimodal histograms. Pore detection uses dual thresholding with a Sauvola boundary mask and a percentile seed mask, then iterative geodesic dilation to reconstruct true pore shapes. Grain size is measured with a random line intercept method and a minimum gap filter, and the pipeline is validated against ImageJ manual segmentation and compared with PoreSpy and GSAT tools.

What would settle it

Hold out a subset of SEM images per temperature class, tune the Sauvola parameters only on the remaining images, and compute IoU on the held-out set against manual segmentations; if the held-out IoU drops substantially below the reported 92.18% and 99.35% means, the claimed agreement with expert analysis is an artifact of parameter tuning rather than a general property of the workflow.

Watch

Extended reading notes

Core claim

The central claim is that a combination of topological noise filtering and local adaptive thresholding is sufficient to segment microstructures whose phases cannot be separated by global intensity thresholds. Sauvola thresholding computes a per-pixel threshold from the local mean and standard deviation, which handles overlapping intensity ranges, while PerSplat removes noise by topological persistence and preserves significant features. The paper reports IoU values against manual segmentations of 92.18% mean at 1200°C and 99.35% mean at 1400°C (with maxima 95.14% and 99.85% cited in the abstract), and shows that the pipeline's porosity, grain size, and pore size measurements are closer to manual reference values than the PoreSpy and GSAT comparison tools. The conclusion states that the workflow is in agreement with expert manual segmentation results.

Load-bearing premise

The evaluation assumes that the manual segmentations used to score the pipeline are independent of the manual segmentations used to tune the Sauvola parameters, but the paper states the parameters were tuned per temperature class to compare against manually thresholded images, so the IoU scores may reflect fitting rather than generalization.

Editorial extensions

If this is right

  • Materials labs can generate annotated microstructure datasets at high throughput, removing a major bottleneck in training machine learning models for processing-structure-property relationships.
  • The same local-thresholding-plus-topological-filtering approach could be applied to other material systems whose phases overlap in image intensity, not just sintered alumina.
  • The per-pore and per-grain measurements the pipeline produces enable statistical comparison of microstructures across processing conditions such as sintering temperature.
  • The workflow's speed makes it practical to analyze large image stacks, supporting future three-dimensional characterization from tomographic data.
  • The authors state a next step of conditional generative models for microstructures, which would rely on datasets this pipeline can produce.

Reading between the lines

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

  • The reported IoU values depend on per-temperature-class tuning of the Sauvola parameters k and w, so a reader should test the pipeline on held-out images before treating the agreement with manual analysis as a general capability.
  • Validation on only two sintering temperatures of one material means the workflow's transferability to other ceramics, magnifications, detectors, or charging conditions remains an open question.
  • The gap between the mean and maximum IoU in the results suggests some images are substantially harder; reporting per-image failure modes could help target future improvements.
  • The grain size portion of the workflow still relies on several hand-set thresholds and gap filters, so the pipeline automates segmentation but not every step of characterization.
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

4 major / 5 minor

Summary. The manuscript presents an automated image-analysis workflow for extracting porosity, grain size, and pore size distributions from SEM images of sintered alumina. The workflow combines noise filtering (selecting PerSplat), Sauvola local thresholding, morphological pore reconstruction, and intercept-based grain size measurement. The authors validate their pipeline against manual ImageJ segmentations for samples sintered at 1200°C and 1400°C, reporting Intersection over Union (IoU) scores of 95.14% and 99.85% in the Abstract, and conclude that the workflow is in agreement with expert manual segmentation. Additional comparisons are made with NIST GSAT for grain size and PoreSpy for pore properties. The central claim is that the automated pipeline can replace manual microstructure characterization for generating materials datasets.

Significance. If the reported agreement with manual segmentation were valid, this workflow would be a useful contribution to automated microstructure characterization, an area of active interest for materials informatics. The paper draws on established open-source tools, provides a documented five-phase pipeline, and attempts quantitative comparisons against three reference methods (manual ImageJ analysis, NIST GSAT, and PoreSpy). The strengths are the reproducible use of standard image-processing libraries, the explicit description of hyperparameters, and the clear presentation of the application context. However, the validation is undermined by a mathematical inconsistency between the reported IoU values and the reported porosity values, and by a circular evaluation protocol in which the same manual segmentations are used both to tune the parameters and to score the results. These issues directly affect the central claim and prevent the paper, in its current form, from substantiating its stated contribution.

major comments (4)
  1. [Section 3.2, Tables 1 and 3] The reported IoU values for 1400°C (mean 99.35%, min 98.64%) are mathematically incompatible with the reported porosity values. With the workflow porosity of 2.44±0.68% (Table 3) and the manual porosity of 0.27±0.19% (Table 1), the maximum possible IoU is min(A_pred,A_true)/max(A_pred,A_true): about 11% if the positive class is pores, and at most (1−0.0244)/(1−0.0027) ≈ 97.8% if the positive class is solid. Neither bound reaches 98.64%, so the IoU values in Table 2 cannot have been computed from the same segmentation masks that produced the porosity statistics in Tables 1 and 3. This internal inconsistency invalidates the paper's central validation claim.
  2. [Sections 2.4 and 3.1] The validation is circular. Section 2.4 states that the Sauvola parameters k and w were tuned 'for each temperature class to compare against manually thresholded images,' and Section 3.1 states that PerSplat was selected because it 'yielded the porosity value closest to the manual segmentation.' Table 2 then reports IoU scores against those same manual segmentations, with no held-out images or cross-validation. The IoU values therefore reflect fit to the labels used for optimization, not independent agreement with expert segmentation. This does not establish the claimed generalizable agreement in the Abstract and Conclusion.
  3. [Abstract and Table 2] The Abstract reports IoU scores of 95.14% and 99.85%, which are the maximum values in Table 2, not the mean values (92.18% and 99.35%). Reporting the best-case values as the overall validation result overstates the performance and is inconsistent with the summary statistics the authors themselves present.
  4. [Tables 1 and 3] The porosity values for PerSplat in Table 1 (0.03±0.01 for 1200°C and 0.0044±0.0029 for 1400°C, presumably as fractions) are inconsistent with the workflow porosity in Table 3 (12.4±2.7% and 2.44±0.68%). The paper does not explain whether Table 1 reports porosity after filtering alone, after thresholding alone, or with different units; if the values are fractions, 0.03 would be 3%, not 12.4%. This inconsistency makes it impossible to interpret the filter comparison and further undermines the validation narrative.
minor comments (5)
  1. [Section 3.3] There is a typo: 'minium gap filter' should be 'minimum gap filter.'
  2. [Conclusion] The Conclusion refers to 'validating the synthetic data generation,' but this paper does not describe any synthetic data generation; this phrase appears to be a leftover from a different context and should be corrected.
  3. [Sections 2.5 and 3.3] The grain-size analysis uses several per-class free parameters (binary map thresholds 20 and 35, minimum gap filters of 3 and 12 pixels) whose selection process is not described; please clarify whether these were also tuned against the manual segmentations, since this would extend the circularity concern.
  4. [Table 1] The table would be clearer if it stated explicitly whether porosity values are fractions or percentages, since the manual row mixes a fraction (0.09) with the corresponding percentage (9%) in later tables.
  5. [Introduction, reference [20]] The text cites 'Choi and Choi' in the context of watershed segmentation, but reference [20] in the bibliography is Heilbronner, R., 'Automatic grain boundary detection and grain size analysis...'; the citation appears mismatched.

Circularity Check

3 steps flagged · score 8.0 of 10

Validation reduces to fitting: Sauvola hyperparameters and PerSplat filter selection are tuned/selected against the same manual segmentations later reported as the IoU and porosity reference.

  1. fitted input called prediction [Section 2.4, Image Binarization, after Eq. (1)]
    "Local hyperparameter optimization was performed, tuning the k parameter and window size w for each temperature class to compare against manually thresholded images."

    The Sauvola threshold parameters k and w are optimized per temperature class against the manual thresholded images, and Table 2 then reports IoU between the automated Sauvola output and those same manual segmentations. The IoU in Eq. (6) is therefore a fit-to-reference score on the tuning set, not an independent validation. The conclusion of agreement with expert manual segmentation follows from the tuned parameters rather than from a held-out prediction.

  2. fitted input called prediction [Section 3.1, Noise Filtering Performance, Table 1 discussion]
    "PerSplat performed well with an SNR increase of 25.0 dB for 1200°C and 22.3 dB for 1400°C images, yielding the porosity value closest to the manual segmentation."

    PerSplat is selected as the pipeline's noise filter because it gives a porosity value closest to the manual segmentation. The same manual porosity values are later used in Table 3 as the reference for validating the workflow's porosity. Thus the filter-selection criterion and the validation target are identical, so the reported closeness of the final porosity to manual values is a selection artifact rather than an independent check.

1 more flagged steps
  1. other [Section 4, Conclusion]
    "The key outcome of this study was the establishment of a workflow that is in agreement with expert manual segmentation results."

    This conclusion is supported only by Table 2 IoU scores and Table 3 porosity comparisons, both of which use manual segmentations that were also used to tune the Sauvola parameters and to select PerSplat. The agreement asserted in the Conclusion is therefore a restatement of fit-to-reference rather than an independent validation result.

full rationale

The paper's central validation is not an independent test. Section 2.4 explicitly states that k and w were tuned per temperature class 'to compare against manually thresholded images', and Table 2 then reports IoU against those same manual thresholded images, making the reported IoU a training-set fit. Section 3.1 selects PerSplat because it yields the porosity value closest to the manual segmentation, and Table 3 later uses that same manual porosity as the validation reference, making the filter-selection criterion identical to the validation target. The Conclusion's claim of agreement with expert manual segmentation therefore follows by construction from these tuning and selection steps. No held-out set, independent ground truth, or machine-checked reproduction is provided. Separately, the reported numbers are internally inconsistent: the 1400°C mean IoU of 99.35% exceeds the theoretical Jaccard bound (about 97.8%) implied by the reported workflow and manual porosity values (2.44% vs 0.27%) under the solid-label reading, and the abstract quotes the Table 2 maximum values (99.85%, 95.14%) rather than the means. These issues compound the circularity but are not necessary to establish it; the explicit tuning and selection language alone shows that the validation is a fit to the reference, not a prediction.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The pipeline depends on several hand-chosen or fitted parameters, most importantly the Sauvola k and w that were tuned against the same manual labels used for validation. It also assumes the manual segmentation is ground truth and that pores are the darkest phase. No new physical entities are introduced.

free parameters (8)
  • Sauvola sensitivity k = not reported
    Tuned per temperature class against manually thresholded images (Section 2.4).
  • Sauvola window size w = not reported
    Tuned per temperature class against manual segmentation (Section 2.4).
  • Grain boundary binary threshold = 20 (1200°C), 35 (1400°C)
    Chosen by hand in the grain size workflow (Section 2.5).
  • Minimum gap filter for intercept counting = 3 pixels (1200°C), 12 pixels (1400°C)
    Chosen per class to filter spurious boundary crossings (Section 2.5).
  • CLAHE clip limit = 2.0
    Chosen in pore detection preprocessing (Section 2.5).
  • CLAHE tile size = 8x8 pixels
    Chosen in pore detection preprocessing (Section 2.5).
  • Percentile seed threshold = 5th percentile of image intensity
    Used as seed mask for morphological pore reconstruction (Section 2.5).
  • Number of random measurement lines = 50 per image
    Chosen based on recommendations for lognormal grain size distributions (Section 2.5); affects the variance of grain size estimates.
assumptions (4)
  • domain assumption Manual segmentation in ImageJ is an accurate ground truth.
    The entire validation rests on comparing automated results to manual analyses (Sections 2.4, 3.2). No independent physical measurement is used to arbitrate.
  • domain assumption Pores appear darker than grains in the SEM images.
    The pore seed mask takes pixels below the 5th intensity percentile as definite pores (Section 2.5), assuming the pore phase is the darkest phase.
  • domain assumption 2D planar sections adequately represent 3D microstructure for grain and pore size statistics.
    The intercept method and pore area measurements are performed on single SEM cross-sections without stereological corrections (Section 2.5).
  • standard math Standard image-processing libraries (OpenCV, scikit-image, PoreSpy) implement the described operations correctly.
    No code is shipped; the pipeline relies on the correctness of third-party implementations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Computational Microstructure Analysis of Sintered Ceramics." pith.science (2026). https://pith.science/paper/XQK3OJZO

@misc{pith2026260804216,
  author       = {Pith},
  title        = {Pith review of: Computational Microstructure Analysis of Sintered Ceramics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XQK3OJZO}},
  note         = {Machine review of arXiv:2608.04216}
}
abstract

Characterizing materials through manual extraction of physical properties from microstructure images is a laborious process. This work presents a workflow to extract porosity, solid fraction, grain size distribution, and pore size distribution from scanning electron microscopy (SEM) images of sintered ceramic samples using an automated pipeline. The primary challenge for extracting physical properties from SEM images is the presence of unimodal histograms in SEM images as a result of the overlapping intensity ranges for the grain and pore phases. We evaluated several different methods for noise reduction and local thresholding of SEM images. We find that topological filtering in combination with Sauvola thresholding enables segmentation and extraction of physical property data from SEM images. We validated the automated pipeline by comparing our results with the results of manual analyses performed for samples sintered at 1200$^o$C and 1400$^o$C and achieved an Intersection over Union (IoU) score of 95.14% and 99.85%, respectively. The workflow provides an efficient means to automatically extract microstructure properties from SEM images as a crucial step in generating materials datasets for machine learning.

Figures

Figures reproduced from arXiv: 2608.04216 by the authors.

Figure 1
Figure 1. Examples of SEM Images of slip-cast alumina samples sintered at (left) 1200◦C and (right) 1400◦C. 2.2 Image Analysis Pipeline The Python programming language and standard libraries that support scientific calculations are the basis for developing the automation pipeline for the project. These include image processing library, OpenCV [7], the scientific computation [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Comparison of noise filtering algorithms. has already been used successfully on low-contrast biomedical images [32] and metallography [9]. Sauvola thresholding computes local threshold as T(x, y) = µ(x, y)  1 + k  σ(x, y) R − 1  , (1) where x and y are the pixel coordinates, µ(x, y) and σ(x, y) are the local mean and standard deviation computed over a square window size w × w centered at pixel (x, y), k is a sen… view at source ↗
Figure 3
Figure 3. Grain size analysis workflow: (top-left) original image, (top-right) morphologi￾cal gradient, (bottom-left) detected boundaries, and (bottom-right) measurement lines with edge points. Pore detection utilizes morphological reconstruction using dual threshold￾ing [14]. This includes two complementary masks, the Sauvola threshold to de￾tect all pore boundaries, and a percentile threshold that detects definite pore regi… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Processing steps for computation of pore size distribution: (top row, left to right) Original SEM image, preprocessed image with bilateral filtering and CLAHE, final pore mask; (bottom row) pore overlay showing detected pores, and individual pores color-coded for analy…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 31 canonical work pages

  1. [1]

    Metallography4(1), 59–78 (Feb 1971)

    Abrams, H.: Grain size measurement by the intercept method. Metallography4(1), 59–78 (Feb 1971). https://doi.org/10.1016/0026-0800(71)90005-X

    work page doi:10.1016/0026-0800(71)90005-x 1971

  2. [2]

    IEEE Transactions on Pattern Analysis and Machine Intelligence16(6), 641–647 (Jun 1994)

    Adams, R., Bischof, L.: Seeded region growing. IEEE Transactions on Pattern Analysis and Machine Intelligence16(6), 641–647 (Jun 1994). https://doi.org/10. 1109/34.295913

    work page 1994

  3. [3]

    MRS Communications9(3), 779–792 (Sep 2019)

    Agrawal, A., Choudhary, A.: Deep materials informatics: Applications of deep learning in materials science. MRS Communications9(3), 779–792 (Sep 2019). https://doi.org/10.1557/mrc.2019.73

    doi:10.1557/mrc.2019.73 2019

  4. [4]

    - References - Scientific Research Pub- lishing, https://www.scirp.org/reference/referencespapers?referenceid=3247763

    ASTME112-13(2013)StandardTestMethodsforDeterminingAverageGrainSize, ASTM International, West Conshohocken. - References - Scientific Research Pub- lishing, https://www.scirp.org/reference/referencespapers?referenceid=3247763

    work page 2013

  5. [5]

    Scientific Reports8(1), 2128 (Feb 2018)

    Azimi, S.M., Britz, D., Engstler, M., Fritz, M., Mücklich, F.: Advanced Steel Mi- crostructural Classification by Deep Learning Methods. Scientific Reports8(1), 2128 (Feb 2018). https://doi.org/10.1038/s41598-018-20037-5, publisher: Nature Publishing Group

    work page doi:10.1038/s41598-018-20037-5 2018

  6. [6]

    Dynamic Thresholding of Gray Level Image,

    J. Bernsen, “Dynamic Thresholding of Gray Level Image,” ICPR‘86 Proceedings of International Conference on Pattern Recognition, Berlin, 1986, pp. 1251-1255. - References - Scientific Research Publishing

    work page 1986

  7. [7]

    Bradski, G.: The Opencv Library, vol. 25. Dr. Dobb’s Journal of Software Tools (Nov 2000), journal Abbreviation: Dr. Dobb’s J. Softw. Tools Publication Title: Dr. Dobb’s J. Softw. Tools 12 N. Dhanankam et al

    work page 2000

  8. [8]

    In: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recog- nition (CVPR’05)

    Buades, A., Coll, B., Morel, J.M.: A non-local algorithm for image denoising. In: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recog- nition (CVPR’05). vol. 2, pp. 60–65 vol. 2 (Jun 2005). https://doi.org/10.1109/ CVPR.2005.38, iSSN: 1063-6919

    work page 2005

Show all 44 references
  1. [9]

    Computational Materials Science123, 176–187 (Oct 2016)

    Chowdhury, A., Kautz, E., Yener, B., Lewis, D.: Image driven machine learning methods for microstructure recognition. Computational Materials Science123, 176–187 (Oct 2016). https://doi.org/10.1016/j.commatsci.2016.05.034

    2016 doi

  2. [10]

    Csurka, G., Larlus, D., Perronnin, F.: What is a good evaluation measure for semantic segmentation? In: Procedings of the British Machine Vision Conference

  3. [11]

    Journal of the European Ceramic Society43(Jul 2023)

    Dadkhah, M., Tulliani, J.M., Saboori, A., Iuliano, L.: Additive Manufacturing of Ceramics: Advances, Challenges, and Outlook. Journal of the European Ceramic Society43(Jul 2023). https://doi.org/10.1016/j.jeurceramsoc.2023.07.033

    2023 doi

  4. [12]

    Wear249, 482–488 (Jun 2001)

    Decencière, E., Jeulin, D.: Morphological decomposition of the surface topography of an internal combustion engine cylinder to characterize wear. Wear249, 482–488 (Jun 2001). https://doi.org/10.1016/S0043-1648(01)00579-8

    2001 doi

  5. [13]

    Microscopy and Microanalysis25(1), 21–29 (Feb 2019)

    DeCost, B.L., Lei, B., Francis, T., Holm, E.A.: High Throughput Quantitative Metallography for Complex Microstructures Using Deep Learning: A Case Study in Ultrahigh Carbon Steel. Microscopy and Microanalysis25(1), 21–29 (Feb 2019). https://doi.org/10.1017/S1431927618015635

    2019 doi

  6. [14]

    Image Analysis and Stereology27(2), 79– 85 (2008)

    Girard, E., Chaix, J.M., Valdivieso, F., Goeuriot, P., Lechelle, J.: LOCAL STUDY OF DEFECTS DURING SINTERING OF UO2: IMAGE PROCESSING AND QUANTITATIVE ANALYSIS TOOLS. Image Analysis and Stereology27(2), 79– 85 (2008). https://doi.org/10.5566/ias.v27.p79-85

    2008 doi

  7. [15]

    In: Goldstein, J.I., Newbury, D.E., Michael, J.R., Ritchie, N.W., Scott, J.H.J., Joy, D.C

    Goldstein, J.I., Newbury, D.E., Michael, J.R., Ritchie, N.W.M., Scott, J.H.J., Joy, D.C.: SEM Image Interpretation. In: Goldstein, J.I., Newbury, D.E., Michael, J.R., Ritchie, N.W., Scott, J.H.J., Joy, D.C. (eds.) Scanning Electron Microscopy and X-Ray Microanalysis, pp. 111–1...

    2018 doi

  8. [16]

    and Woods, R.E

    Gonzalez, R.C. and Woods, R.E. (2018) Digital Image Processing. 4th Edition, Pearson Education, London. - References - Scientific Research Publishing

    2018

  9. [17]

    Journal of Open Source Software4, 1296 (May 2019)

    Gostick, J., Khan, Z.A., Tranter, T., Kok, M., AGNAOU, M., Sadeghi, A., Jervis, R.: PoreSpy: A Python Toolkit for Quantitative Analysis of Porous Media Images. Journal of Open Source Software4, 1296 (May 2019). https://doi.org/10.21105/ joss.01296

    2019

  10. [18]

    Gostick, J.T.: Versatile and efficient pore network extraction method using marker- basedwatershedsegmentation.PhysicalReviewE96(2),023307(Aug2017).https: //doi.org/10.1103/PhysRevE.96.023307, publisher: American Physical Society

    doi

  11. [19]

    In: Image Analysis and Recognition

    Hadjadj, Z., Meziane, A., Cherfa, Y., Cheriet, M., Setitra, I.: ISauvola: Improved Sauvola’s algorithm for document image binarization. In: Image Analysis and Recognition. ICIAR 2016. Lecture Notes in Computer Science, vol. 9730, pp. 737–

    2016

  12. [20]

    Journal of Structural Geology22, 969–981 (Jul 2000)

    Heilbronner, R.: Automatic grain boundary detection and grain size analysis using polarization micrographs or orientation images. Journal of Structural Geology22, 969–981 (Jul 2000). https://doi.org/10.1016/S0191-8141(00)00014-6

    doi

  13. [21]

    IEEE Transactions on Acoustics, Speech, and Signal Processing27(1), 13–18 (Feb 1979)

    Huang, T., Yang, G., Tang, G.: A fast two-dimensional median filtering algorithm. IEEE Transactions on Acoustics, Speech, and Signal Processing27(1), 13–18 (Feb 1979). https://doi.org/10.1109/TASSP.1979.1163188 Computational Microstructure Analysis of Sintered Ceramics 13

    1979

  14. [22]

    New Phytologist 11(2),37–50(1912).https://doi.org/10.1111/j.1469-8137.1912.tb05611.x,_eprint: https://nph.onlinelibrary.wiley.com/doi/pdf/10.1111/j.1469-8137.1912.tb05611.x

    Jaccard, P.: The Distribution of the Flora in the Alpine Zone. New Phytologist 11(2),37–50(1912).https://doi.org/10.1111/j.1469-8137.1912.tb05611.x,_eprint: https://nph.onlinelibrary.wiley.com/doi/pdf/10.1111/j.1469-8137.1912.tb05611.x

    1912

  15. [23]

    Computer Vision, Graph- ics, and Image Processing29(3), 273–285 (Mar 1985)

    Kapur, J.N., Sahoo, P.K., Wong, A.K.C.: A new method for gray-level pic- ture thresholding using the entropy of the histogram. Computer Vision, Graph- ics, and Image Processing29(3), 273–285 (Mar 1985). https://doi.org/10.1016/ 0734-189X(85)90125-2

    1985

  16. [24]

    Journal of Structural Geology135, 104042 (Jun 2020)

    Lopez-Sanchez, M.A.: Which average, how many grains, and how to estimate ro- bust confidence intervals in unimodal grain size populations. Journal of Structural Geology135, 104042 (Jun 2020). https://doi.org/10.1016/j.jsg.2020.104042

    2020

  17. [25]

    Journal of Visual Communi- cation and Image Representation1(1), 21–46 (Sep 1990)

    Meyer, F., Beucher, S.: Morphological segmentation. Journal of Visual Communi- cation and Image Representation1(1), 21–46 (Sep 1990). https://doi.org/10.1016/ 1047-3203(90)90014-M

    1990

  18. [26]

    usnistgov/grain-size-analysis-tools (Feb 2022), https://github.com/usnistgov/ grain-size-analysis-tools, original-date: 2024-07-10T21:47:29Z

    2022

  19. [27]

    Journal of the American Ceramic Society91(12), 4059–4067 (2008)

    Münch, B., Holzer, L.: Contradicting Geometrical Concepts in Pore Size Analysis Attained with Electron Microscopy and Mer- cury Intrusion. Journal of the American Ceramic Society91(12), 4059–4067 (2008). https://doi.org/10.1111/j.1551-2916.2008.02736.x, _eprint: https://cerami...

    2008

  20. [28]

    Englewood Cliffs, N.J

    Niblack, W.: An introduction to digital image processing. Englewood Cliffs, N.J. : Prentice-Hall International (1986)

    1986

  21. [29]

    IEEE Transactions on Systems, Man, and Cybernetics9(1), 62–66 (Jan 1979)

    Otsu, N.: A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics9(1), 62–66 (Jan 1979). https: //doi.org/10.1109/TSMC.1979.4310076

    1979

  22. [30]

    Computational Materials Science 202, 110920 (2022) - Google Search

    Patel, A., Hou, T., Rodriguez, J., Dey, T., Birnie, D.: Patel, A.V., Hou, T., Ro- driguez, J.D.B., Dey, T.K., Birnie, D.P.: Topological filtering for 3D microstruc- ture segmentation. Computational Materials Science 202, 110920 (2022) - Google Search. https://doi.org/10.48550/...

    2022 doi arXiv

  23. [31]

    In: Proceedings of the International Symposium on Visual Computing

    Rahman, M.A., Wang, Y.: Optimizing Intersection-Over-Union in Deep Neural Networks for Image Segmentation. In: Proceedings of the International Symposium on Visual Computing. vol. 10072, pp. 234–244 (Dec 2016). https://doi.org/10. 1007/978-3-319-50835-1_22

    2016

  24. [32]

    https://doi.org/10.48550/arXiv.1505.04597, arXiv:1505.04597 [cs]

    Ronneberger, O., Fischer, P., Brox, T.: U-Net: Convolutional Networks for Biomed- ical Image Segmentation (May 2015). https://doi.org/10.48550/arXiv.1505.04597, arXiv:1505.04597 [cs]

    2015 doi arXiv

  25. [33]

    Physica D: Nonlinear Phenomena60(1), 259–268 (Nov 1992)

    Rudin, L.I., Osher, S., Fatemi, E.: Nonlinear total variation based noise removal algorithms. Physica D: Nonlinear Phenomena60(1), 259–268 (Nov 1992). https: //doi.org/10.1016/0167-2789(92)90242-F

    1992 doi

  26. [34]

    Pattern Recognition33(2), 225–236 (Feb 2000)

    Sauvola, J., Pietikäinen, M.: Adaptive document image binarization. Pattern Recognition33(2), 225–236 (Feb 2000). https://doi.org/10.1016/S0031-3203(99) 00055-2

    2000 doi

  27. [35]

    npj Computa- tional Materials5(1), 83 (Aug 2019)

    Schmidt, J., Marques, M.R.G., Botti, S., Marques, M.A.L.: Recent advances and applications of machine learning in solid-state materials science. npj Computa- tional Materials5(1), 83 (Aug 2019). https://doi.org/10.1038/s41524-019-0221-0, publisher: Nature Publishing Group

    2019 doi

  28. [36]

    Protein Science : A Publication of the Protein Society30(1), 234–249 (Jan 2021)

    Schroeder, A.B., Dobson, E.T.A., Rueden, C.T., Tomancak, P., Jug, F., Eliceiri, K.W.: The ImageJ ecosystem: Open-source software for image visualization, pro- cessing, and analysis. Protein Science : A Publication of the Protein Society30(1), 234–249 (Jan 2021). https://doi.or...

    2021 doi

  29. [37]

    Journal of Electronic Imaging13, 146–168 (Jan 2004)

    Sezgin, M., Sankur, B.: Survey over image thresholding techniques and quantitative performance evaluation. Journal of Electronic Imaging13, 146–168 (Jan 2004). https://doi.org/10.1117/1.1631315

    2004 doi

  30. [38]

    https://doi.org/10.5555/773286

    Morphological Image Analysis: Principles and Applications | Guide books | ACM Digital Library. https://doi.org/10.5555/773286

    doi

  31. [39]

    In: Sixth International Conference on Computer Vision

    Tomasi, C., Manduchi, R.: Bilateral filtering for gray and color images. In: Sixth International Conference on Computer Vision. pp. 839–846 (Jan 1998). https:// doi.org/10.1109/ICCV.1998.710815

    1998

  32. [40]

    IEEE Transactions on Image Processing2(2), 176–201 (Apr 1993)

    Vincent, L.: Morphological grayscale reconstruction in image analysis: applications and efficient algorithms. IEEE Transactions on Image Processing2(2), 176–201 (Apr 1993). https://doi.org/10.1109/83.217222

    1993 doi

  33. [41]

    PeerJ2, e453 (Jun 2014)

    Walt, S.v.d., Schönberger, J.L., Nunez-Iglesias, J., Boulogne, F., Warner, J.D., Yager, N., Gouillart, E., Yu, T., contributors, t.s.i.: scikit-image: Image process- ing in Python. PeerJ2, e453 (Jun 2014). https://doi.org/10.7717/peerj.453, arXiv:1407.6245 [cs]

    2014 arXiv

  34. [42]

    https://doi

    Contrast limited adaptive histogram equalization | Graphics gems IV. https://doi. org/10.5555/180895.180940

  35. [745]

    https://doi.org/10.1007/978-3-319-41501-7_82

    Springer, Cham (2016). https://doi.org/10.1007/978-3-319-41501-7_82

    2016 doi

  36. [2013]

    32.1–32.11

    pp. 32.1–32.11. British Machine Vision Association, Bristol (2013). https: //doi.org/10.5244/C.27.32

    2013 doi

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.