Pith. sign in

REVIEW 3 major objections 5 minor 36 references

Learning morphological operators for skin detection

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A learned morphological post-processing step, applied to the output of any skin detector, improves accuracy on almost all of ten diverse benchmark datasets.

desk verdict A modest but honest adaptive morphology post-processor; the SegNet+TM evidence is real, but the EM threshold sentence must be clarified before the trained-morphology claims hold. read the letter →

arxiv 1908.03630 v2 pith:MGOY6CFE submitted 2019-08-09 cs.CV

Alessandra Lumini , Loris Nanni , Alice Codogno , Filippo Berno This is my paper · ORCID
classification cs.CV
keywords skindetectionsegmentationmorphologicaloperatorspost-processingconvolutionalneuralnetworksSegNetSA3imageclassification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that skin-detection systems can be improved without retraining them, by adding a cheap post-processing stage built from trained morphological operators. The idea is to classify each raw skin mask into one of five image-pattern classes using three simple statistics, then apply a class-specific sequence of erosions, dilations, fillings, and small-object removals that strip away false-positive background pixels. The authors test the post-processor on top of a deep-learning detector (a fine-tuned SegNet) and a handcrafted detector (SA3), across ten public datasets. Their central claim is that trained morphology outperforms a fixed morphological baseline and no post-processing, with the best combination ranking first among all compared approaches.

What carries the argument

The central mechanism is a five-class decision rule plus class-specific morphological operator sequences. For a binary mask $BW$, the method computes the skin ratio $SR$ on a hole-filled copy; if $SR$ exceeds threshold $a_1$, the mask is heavily eroded and the number of connected components $CC$ is counted, and if $CC$ is below $b_1$, the border skin ratio $BSR$ decides between class A and class B. Additional thresholds $a_2$ and $b_2$ separate class D from the default class E. Each class then receives a fixed chain of morphological commands, with the structural element a disk of radius 6, and the final mask is multiplied pixel-wise with the original segmentation. This mechanism is what carries the claimed gain: it converts a one-size-fits-all post-processing into a per-image adaptive cleanup at negligible computational cost.

What would settle it

Freeze the thresholds after training on the first 2000 ECU images only, run the identical SegNet post-processing on all ten datasets, and compare F1/AP with the unprocessed SegNet; if the post-processed version does not beat the raw detector on a clear majority of datasets, the central claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that false-positive errors in skin segmentation are structured enough to be corrected by a small set of learned image-processing rules. Each binary output mask is described by its skin ratio, connected-component count, and border skin ratio; these three features place the mask into classes A through E, and a lookup table then applies a fixed operator sequence per class, with class A (background classified as skin) receiving an extra step that removes the largest connected component. The thresholds for the rules are obtained by grid search, either on the ECU training set or by leave-one-out over the other datasets. In the reported experiments the trained versions (TM and EM) beat the base morphological version (BM), which in turn beats the raw detectors on almost all datasets; the Wilcoxon signed-rank test gives SegNet+TM over SegNet a p-value of 0.0058.

Load-bearing premise

The whole improvement rests on the five grid-search thresholds $(a_1,a_2,b_1,b_2,c_1)$ transferring from the training set to every test dataset; if those thresholds were tuned with any access to test data, or if they do not transfer to a new domain, the reported gains would shrink or vanish.

Editorial extensions

If this is right

  • Any current skin detector can be upgraded by appending this post-processor, with no retraining of the underlying model and a runtime of roughly 0.03 seconds per 224x224 image.
  • The class-based formulation means the thresholds can be re-learned on a small labeled set, so the method should adapt to a new domain cheaply.
  • Gains concentrate on false-positive reduction; detectors that mostly miss skin regions will not be helped, because the post-processor cannot recover false negatives.
  • Statistical support from the Wilcoxon test suggests the improvement over both the raw detector and the fixed morphological baseline is not a single-dataset accident.

Reading between the lines

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

  • Because the post-processor reads only binary masks and three scalar statistics, a natural untested extension is to apply the same scheme to other binary segmentation outputs where false positives from background are the dominant error.
  • The paper itself notes that working from probability masks instead of binary masks could recover false negatives; a testable follow-up would feed soft masks through the same class rules with several thresholds.
  • The reported sentence that optimal EM parameters were learned 'using all the datasets' leaves open whether the comparison is fully held out; a strict holdout replication is the cleanest way to confirm the transfer claim.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a post-processing method for skin detection that applies learned morphological operators to the binary masks produced by an existing skin detector. The method defines five image classes (A through E) based on skin ratio, connected-component count, and border skin ratio, and assigns each class a fixed sequence of morphological operations. Class-assignment thresholds (a1, a2, b1, b2, c1) are tuned by grid search. The authors evaluate the post-processor on two base detectors, SegNet (deep learning) and SA3 (handcrafted), over ten public datasets from their own earlier benchmark, comparing a fixed baseline (BM), a leave-one-out trained variant (TM), and an ECU-trained variant (EM). Results are reported as per-dataset F1/AP scores, a global rank, and Wilcoxon signed-rank tests. The central claim is that the trained post-processor improves both base detectors in almost all datasets and that the trained variants outperform the fixed baseline.

Significance. If the main claim holds, the work offers a cheap, task-agnostic post-processing module that can raise skin-detection accuracy across diverse datasets, which would be practically valuable given the negligible computational cost (about 33 ms for a 224×224 image). The paper has concrete strengths: the evaluation uses ten heterogeneous public datasets; the code is promised open-source; and the Wilcoxon test for SegNet+TM over SegNet (p=0.0058) provides legitimate statistical evidence for the SegNet side of the claim. The leave-one-out protocol for TM is not circular, and the SegNet+TM improvement is a genuinely held-out result. However, the significance is materially weakened by (i) an explicit ambiguity about whether the EM thresholds were selected on the test distribution, and (ii) the absence of any significance test for the SA3 results, where trained morphology actually degrades performance on three of the ten datasets.

major comments (3)
  1. [Section 3, after Table 4] The sentence "The optimal EM parameters (a1,a2,b1,b2,c1) learned for SA3 and SegNet using all the datasets are ..." is directly inconsistent with the preceding description of EM as training on the first 2000 images of ECU. If "using all the datasets" is literal, then the EM columns in Table 4 are the result of selecting thresholds on the test distribution, making the EM-versus-BM comparison and the conclusion that "EM and TM perform similarly" unsupported. This is load-bearing because the paper explicitly concludes that "a simple training on a small set of images is enough to tune our approach," a claim that rests on the EM results. The authors must state unambiguously which images were used to select each threshold set. If EM was tuned on all ten datasets, the EM results must be recomputed under a genuinely held-out training set (e.g., the first 2000 ECU images only) or removed.
  2. [Table 4 and statistical validation] For the SA3 base detector, the TM variant is worse than SA3 alone or SA3+BM on three datasets (Feeval: 0.532 vs 0.539/0.540; SFA: 0.849 vs 0.863/0.849; VMD: 0.140 vs 0.147/0.140), and the paper reports no Wilcoxon test for any SA3 comparison. The claim that "the use of morphological operators improves the performance of base approaches in almost all the datasets" is therefore not statistically established for SA3 as a whole. At minimum, the authors should report the Wilcoxon p-values for SA3+TM vs SA3 and SA3+BM, and discuss whether the improvements on the remaining seven datasets are sufficient to compensate for the degradations, given the small number of datasets.
  3. [Section 2.3 and Table 3] The class definitions and the morphological operator sequences in Table 3 are not fully justified. In particular, the paper states that five classes are defined but then says that classes B, C, and D receive the same operator sequence, so the actual number of distinct post-processing pipelines is three. The text should either explain what distinguishes B, C, and D in terms of the operators (e.g., the later per-component handling for A) or explicitly acknowledge that the class labels mainly drive the background-removal step for class A. As written, the reader cannot determine whether the three-way distinction among B, C, and D affects the output at all.
minor comments (5)
  1. [Abstract] The abstract contains a duplicated phrase: "is performed according to an existing skin detection approach is performed for skin segmentation." This should be corrected to a single clause.
  2. [Figure 1 caption] The caption lists "mask obtained from SA3(τ=50)+Morphological processing" twice; the second instance is presumably intended to be "mask obtained from SegNet+Morphological processing."
  3. [Conclusions] The sentence "Our experiments proved that the performance of SegNet decades for low quality images" contains a typo: "decades" should be "degrades."
  4. [Section 3, performance indicators] The formula for F1 is written as F1 = 2tp/(2tp + fn + fp), which omits the true negatives; this is standard, but the next sentence says "where tn are true negatives" even though tn does not appear in the formula. Either remove tn from the explanation or clarify that tn is not used in the F1 computation.
  5. [Table 1] The class C row lists only SR and CC features and uses "-" for BSR; since the text says "Each rule is intended as the evaluation of the three conditions in AND," the "-" should be explicitly defined as "don't care" to avoid ambiguity.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; EM threshold sentence is a soundness ambiguity, not a circular step.

full rationale

The derivation chain is not circular. The trainable morphology is a post-processor whose thresholds (a1,a2,b1,b2,c1) are obtained by grid search on a training set (Section 2.3: 'Parameter optimization has been performed by grid search on the training set'), with protocols TM (leave-one-out) and EM (ECU first 2000 images); Table 4 then evaluates the resulting operators on the ten benchmark datasets. The base detectors are external (SegNet and SA3 from Kawulok et al.), and the improvement is measured against the original masks of those detectors, so no target quantity is reinserted as an input. The paper's reliance on [1] is a self-citation for choosing the base detectors and benchmark, but the datasets are public and the comparisons are performed in this paper against untouched baseline outputs; the self-citation is not load-bearing. One passage needs clarification but is not a circularity: 'The optimal EM parameters (a1,a2,b1,b2,c1) learned for SA3 and SegNet using all the datasets are ...' contradicts the earlier definition of EM as training on the first 2000 ECU images. If literal, the EM columns in Table 4 would be in-sample threshold fits rather than predictions, which is a data-leakage/soundness concern. However, the TM leave-one-out protocol and the Wilcoxon tests (e.g., SegNet+TM vs SegNet, p=0.0058) are unaffected and provide independent non-circular evidence for the central claim. Score 2 reflects the minor, non-load-bearing self-citations.

Assumptions & free parameters 8 free parameters · 3 assumptions · 0 invented entities

The only genuinely learned quantities are the five grid-searched thresholds; everything else is standard image-processing machinery. The method itself is an ad hoc rule set rather than derived from a model of skin image formation.

free parameters (8)
  • a1 (skin ratio threshold) = 0.3 (SA3 and SegNet)
    Lower bound for SR to consider classes A/B/C; set by grid search.
  • a2 (skin ratio low threshold) = 0.06 (both)
    Lower bound for SR to consider class D; grid-searched.
  • b1 (connected components threshold) = 16 (SA3), 10 (SegNet)
    Upper bound for CC to separate A/B from C; grid-searched.
  • b2 (connected components threshold for D) = 48 (SA3), 40 (SegNet)
    Grid-searched.
  • c1 (border skin ratio threshold) = 0.55 (SA3), 0.25 (SegNet)
    Grid-searched to separate A from B.
  • SA3 tau = 50
    Acceptance threshold of SA3, taken as best from [1].
  • morphological structuring element sizes = disk radius 6; erosion 10, dilation 8 in baseline
    Chosen by hand; not optimized.
  • bwareaopen minimum object size P = not stated exactly; text mentions removing 'cc smaller than 10px' in one place
    A hand-set parameter used to remove small components.
assumptions (3)
  • domain assumption Morphological post-processing can reduce false positives without excessively penalizing true positives.
    Stated in Section 2.3: the method is designed to reduce FP, but it cannot handle false negatives.
  • ad hoc to paper The three features (SR, CC, BSR) computed on binary masks are sufficient to classify images into five post-processing-relevant classes.
    Table 1 defines the rules; no principled derivation is given.
  • domain assumption The testing framework of [1] (same authors) is a fair, standard benchmark for comparing skin detectors.
    The whole evaluation relies on the protocol and dataset split from their previous paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning morphological operators for skin detection." pith.science (2026). https://pith.science/paper/MGOY6CFE

@misc{pith2026190803630,
  author       = {Pith},
  title        = {Pith review of: Learning morphological operators for skin detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MGOY6CFE}},
  note         = {Machine review of arXiv:1908.03630}
}
read the original abstract

In this work we propose a novel post processing approach for skin detectors based on trained morphological operators. The first step, consisting in skin segmentation is performed according to an existing skin detection approach is performed for skin segmentation, then a second step is carried out consisting in the application of a set of morphological operators to refine the resulting mask. Extensive experimental evaluation performed considering two different detection approaches (one based on deep learning and a handcrafted one) carried on 10 different datasets confirms the quality of the proposed method.

Figures

Figures reproduced from arXiv: 1908.03630 by the authors.

Figure 1
Figure 1. Processing of two sample images from the Schmugge dataset. From left to right and from top to bottom: a sample image, its ground truth mask, mask obtained from SA3(τ=50), mask obtained from SA3(τ=50)+Morphological processing, mask obtained from SegNet, mask obtained from SA3(τ=50)+Morphological processing. 2.3. Morphological Processing Analyzing the results regarding the number of true positives (TP) and false posit… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 28 canonical work pages

  1. [25]

    S. Kang, B. Choi, D. Jo, Faces detection method based on skin color modeling, J. Syst. Archit. 64 (2016) 100–109. doi:10.1016/J.SYSARC.2015.11.009

    work page doi:10.1016/j.sysarc.2015.11.009 2016

  2. [1]

    Lumini, L

    A. Lumini, L. Nanni, Fair comparison of skin detection approaches on publicly available datasets, (2018)

    work page 2018

  3. [2]

    Naji, H.A

    S. Naji, H.A. Jalab, S.A. Kareem, A survey on skin detection in colored images, Artif. Intell. Rev. (2018). doi:10.1007/s10462-018-9664-9. First Author et al

    work page doi:10.1007/s10462-018-9664-9 2018

  4. [3]

    Kakumanu, S

    P. Kakumanu, S. Makrogiannis, N. Bourbakis, A survey of skin -color modeling and detection methods, Pattern Recognit. 40 (2007) 1106–1122. doi:10.1016/j.patcog.2006.06.010

    work page doi:10.1016/j.patcog.2006.06.010 2007

  5. [4]

    M.J. Seow, D. Valaparla, V.K. Asari, Neural network based skin color model for face detection, in: Proc. - Appl. Imag. Pattern Recognit. Work., 2004. doi:10.1109/AIPR.2003.1284262

    arXiv 2004

  6. [5]

    R. Khan, A. Hanbury, J. Stoettinger, Skin detection: A random forest approach, in: Proc. - Int. Conf. Image Process. ICIP, 2010. doi:10.1109/ICIP.2010.5651638

    work page arXiv 2010

  7. [6]

    N. Sebe, I. Cohen, T.S. Huang, T. Gevers, Skin detection: a Bayesian network approach, Proc. 17th Int. Conf. Pattern Recognit. 2 (2004) 2–5. doi:10.1109/ICPR.2004.1334405

    arXiv 2004

  8. [7]

    Poudel, J.J

    R.P.K. Poudel, J.J. Zhang, D. Liu, H. Nait -Charif, Skin Color Detection Using Region -Based Approach, Int. J. Image Process. 7 (2013) 385

    work page 2013

Show all 36 references
  1. [8]

    Chen, M.S

    W.C. Chen, M.S. Wang, Region-based and content adaptive skin detection in color images, Int. J. Pattern Recognit. Artif. Intell. 21 (2007) 831–853. doi:10.1142/s0218001407005715

    2007 doi

  2. [9]

    T. Xu, Z. Zhang, Y. Wan g, Patch -wise skin segmentation of human body parts via deep neural networks, J. Electron. Imaging. 24 (2015) 043009. doi:10.1117/1.JEI.24.4.043009

    2015 doi

  3. [10]

    H. Zuo, H. Fan, E. Blasch, H. Ling, Combining Convolutional and Recurrent Neural Networks for Human Sk in Detection, IEEE Signal Process. Lett. 24 (2017) 289–293

    2017

  4. [11]

    Y. Kim, I. Hwang, N.I. Cho, Convolutional neural networks and training strategies for skin detection, in: IEEE Int. Conf. Image Process., 2017: pp. 3919–3923. doi:10.1109/ICIP.2017.8297017

    2017

  5. [12]

    Dourado, F

    A. Dourado, F. Guth, T.E. de Campos, W. Li, Domain adaptation for holistic skin detection, CoRR. abs/1903.0 (2019). http://arxiv.org/abs/1903.06969

    2019 arXiv

  6. [13]

    Dourado, F

    A. Dourado, F. Guth, T.E. de Campos, W. Li, Domain adaptation for holistic skin detection, CoRR. a bs/1903.0 (2019)

    2019

  7. [14]

    W. Song, N. Zheng, R. Zheng, X.B. Zhao, A. Wang, Digital image semantic segmentation algorithms: A survey, J. Inf. Hiding Multimed. Signal Process. 10 (2019) 196–211

    2019

  8. [15]

    Shelhamer, J

    E. Shelhamer, J. Long, T. Darrell, Fully Convolutional Networks for Semantic Segmentation, IEEE Trans. Pattern Anal. Mach. Intell. (2017). doi:10.1109/TPAMI.2016.2572683

    2017

  9. [16]

    Badrinarayanan, A

    V. Badrinarayanan, A. Kendall, R. Cipolla, SegNet: A Deep Convolutional Encoder -Decoder Architecture for Image Segmentation, IEEE Trans. Pattern Anal. Mach. Intell. (2017). doi:10.1109/TPAMI.2016.2644615

    2017

  10. [17]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: Lect. Notes Comput. Sci. (Including Subser. Lect. Notes Artif. Intell. Lect. Note s Bioinformatics), 2015. doi:10.1007/978-3-319-24574-4_28

    2015 doi

  11. [18]

    Arsalan, D.S

    M. Arsalan, D.S. Kim, M. Owais, K.R. Park, OR -Skip-Net: Outer residual skip network for skin segmentation in non-ideal situations, Expert Syst. Appl. 141 (2020) 112922. doi:10.1016/J.ESWA.2019 .112922

    2020 doi

  12. [19]

    Phung, A

    S.L. Phung, A. Bouzerdoum, D. Chai, Skin segmentation using color pixel classification: Analysis and comparison, IEEE Trans. Pattern Anal. Mach. Intell. 27 (2005) 148–154. doi:10.1109/TPAMI.2005.17

    2005 doi

  13. [20]

    Sandnes, L

    F.E. Sandnes, L. Neyse, Y.-P. Huang, Simple and practical skin detection with static RGB -color lookup tables: A visualization-based study, in: IEEE Int. Conf. Syst. Man, Cybern., 2016: pp. 2370–2375

    2016

  14. [21]

    Kawulok, J

    M. Kawulok, J. Kawulok, J. Nalepa, B. Smolka, Self -adaptive algorithm for segmenting skin regions, EURASIP J. Adv. Signal Process. (2014) 1–22. doi:10.1186/1687-6180-2014-170

    2014 doi

  15. [22]

    Hassan, A.R

    E. Hassan, A.R. Hilal, O. Basir, Using ga to optimize the explicitly defined skin regions for human skincolor detection, in: Can. Conf. Electr. Comput. Eng., 2017. doi:10.1109/CCECE.2017.7946699

    2017

  16. [23]

    Gonzalez, R.E

    R.C.. Gonzalez, R.E. Woods, Digital image processing, 2008. doi:10.1049/ep.1978.0474

    2008

  17. [24]

    Yun, H.J

    J.U. Yun, H.J. Lee, A.K. Paul, J.H. Baek, Face detection for video summary using illumination-compensation and First Author et al. morphological processing, Pattern Recognit. Lett. (2009). doi:10.1016/j.patrec.2009.04.010

    2009 doi

  18. [26]

    X. Wang, L. Wang, T. Lei, C. Wang, Face detection based on improved skin model and local iterated conditional modes, in: Proc. - Int. Conf. Nat. Comput., 2016. doi:10.1109/ICNC.2015.7378122

    2016

  19. [27]

    Jones, J.M

    M.J. Jones, J.M. Rehg, Statistical color models with application to skin detectio n, Int. J. Comput. Vis. 46 (2002) 81–96. doi:10.1023/A:1013200319198

    2002 doi

  20. [28]

    Ruiz-Del-Solar, R

    J. Ruiz-Del-Solar, R. Verschae, Skin detection using neighborhood information, in: Proc. - Sixth IEEE Int. Conf. Autom. Face Gesture Recognit., 2004: pp. 463–468. doi:10.1109/AFGR.2004.1301576

    2004 arXiv

  21. [29]

    Schmugge, S

    S.J. Schmugge, S. Jayaram, M.C. Shin, L. V. Tsap, Objective evaluation of approaches of skin detection using ROC analysis, Comput. Vis. Image Underst. 108 (2007) 41–51. doi:10.1016/j.cviu.2006.10.009

    2007 doi

  22. [30]

    Stöttinger, A

    J. Stöttinger, A. Hanbury, C. Liensberger, R. Khan, Skin paths for contextual flagging adult videos, in: Lect. Notes Comput. Sci. (Including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), 2009: pp. 303 –314. doi:10.1007/978-3-642-10520-3_28

    2009 doi

  23. [31]

    Sanmiguel, S

    J.C. Sanmiguel, S. Suja, Skin detection by dual maximization of detectors agreement for video monitoring, Pattern Recognit. Lett. 34 (2013) 2102–2109. doi:10.1016/j.patrec.2013.07.016

    2013 doi

  24. [32]

    Casati, D.R

    J.P.B. Casati, D.R. Moraes, E.L.L. Rodrigues, SFA: A human skin image database based on FERET and AR facial images, in: IX Work. Visao Comput. Rio Janeiro, 2013

    2013

  25. [33]

    Tan, C.S

    W.R. Tan, C.S. Chan, P. Yogarajah, J. Condell, A Fusion Approach for Efficient Human Skin Detection, Ind. Informatics, IEEE Trans. 8 (2012) 138–147. doi:10.1109/TII.2011.2172451

    2012

  26. [34]

    Nanni, A

    L. Nanni, A. Lumini, Skin Detection for Reducing False Positive in Face Detection, in: V.M. Petrova (Ed.), Adv. Eng. Res. Vol.16, Nov. Publ. 2017, Nova Publisher, 2017. https://www.novapublishers.com/catalog/product_info.php?products_id=60326&osCsid=

    2017

  27. [35]

    Angelova, Y

    A. Angelova, Y. Abu -Mostafa, P. Perona, Pruning training sets for learning of object categories, in: Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., 2005: pp. 494 –501. doi:10.1109/CVPR.2005.283

    2005 doi

  28. [36]

    Demšar, Statistical Comparisons of Classifiers over Multiple Data Sets, J

    J. Demšar, Statistical Comparisons of Classifiers over Multiple Data Sets, J. Mach. Learn. Res. 7 (2006) 1 –30. doi:10.1016/j.jecp.2010.03.005

    2006 doi

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.