pith. sign in

arxiv: 1907.05595 · v1 · pith:3TYLSCCQnew · submitted 2019-07-12 · 📡 eess.IV · cs.CV

Jointly Adversarial Network to Wavelength Compensation and Dehazing of Underwater Images

Xueyan Ding , Yafei Wang , Yang Yan , Zheng Liang , Zetian Mi , Xianping Fu This is my paper

Pith reviewed 2026-05-24 22:37 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords underwater image enhancementgenerative adversarial networkwavelength compensationdehazingimage formation modeltransmission mapimage synthesis
0
0 comments X

The pith

A network restores underwater images by jointly compensating for wavelength attenuation along the surface-to-object path and scattering along the object-to-camera path.

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

The paper develops a generative adversarial network that embeds a simplified underwater image formation model to address color casts from wavelength-dependent absorption and blurriness from scattering at the same time. Earlier approaches typically modeled only the scattering component through a hazy formation equation. Separate network modules estimate the transmission map via a multi-scale dense encoder-decoder, the wavelength attenuation coefficients, and the background light; these parameters are then combined in the formation model to recover the image, with an added edge-preserving module for detail. A new synthesis procedure creates training pairs that reproduce the color, contrast, and blur of real water types so the network can be trained without large sets of labeled real images.

Core claim

By embedding a simplified underwater formation model into a generative adversarial network, the jointly wavelength compensation and dehazing network estimates the transmission map, wavelength attenuation, and background light through dedicated modules and applies the model to recover the original scene radiance; a multi-scale densely connected encoder-decoder produces the transmission map while an edge-preserving module refines output detail, and training uses a novel synthesis method that simultaneously reproduces the color, contrast, and blurriness of real underwater environments across different water types.

What carries the argument

The jointly wavelength compensation and dehazing network (JWCDN) that embeds the simplified underwater formation model into GAN modules for joint parameter estimation and image recovery.

If this is right

  • Recovered images exhibit improved color balance, contrast, and sharpness on both synthetic and real underwater data.
  • The method produces results that are comparable or superior to prior techniques on standard subjective and objective image-quality metrics.
  • The synthesis procedure supplies training data that enable the network to handle multiple water types without real paired examples.
  • The edge-preserving module supplies additional detail recovery after the formation-model inversion step.

Where Pith is reading between the lines

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

  • The same joint estimation structure could be tested on other scattering media such as fog or smoke where both wavelength-dependent absorption and particle scattering occur.
  • The synthesis method might serve as a data-generation tool for training other underwater vision models that require realistic color and blur variation.
  • Outputs from this recovery step could be fed directly into downstream underwater tasks such as object detection or 3-D reconstruction to measure end-to-end gains.

Load-bearing premise

The novel underwater image synthesis method generates images whose color, contrast, and blurriness accurately match real-world underwater environments across different water types.

What would settle it

A test set of real underwater images with corresponding clear reference photographs on which the method's outputs retain visible color casts or unresolved blur would show that the joint model fails to generalize.

Figures

Figures reproduced from arXiv: 1907.05595 by Xianping Fu, Xueyan Ding, Yafei Wang, Yang Yan, Zetian Mi, Zheng Liang.

Figure 1
Figure 1. Figure 1: Examples of underwater images with different color tones and [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An overview of the proposed method. (a) and (c) show the generator and discriminator of the proposed JWCDN underwater image restoration method. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic diagram of underwater optical imaging. Natural light enters [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Samples of synthesized underwater images with different types and water depths generated from NYU-depth2 dataset [40]. We show synthesized [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: An overview of the proposed multi-scale densely connected transmission map estimation network. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Transmission map estimation results. R, G and B denotes red, green and blue channels respectively. where UICM denotes the colorfulness, UISM denotes the sharpness, and UIConM denotes the contrast measures. The parameters 1, 2 and 3 are weights, whose values are application dependent. In this paper, the values are as follows: 1 = 0.3282, 2 = 0.2953, and 3 = 3.5753. A greater UIQM value indicates super… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparisons for samples from the synthetic test dataset [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparisons for samples from the real-world image dataset [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparisons for samples from real-world image dataset [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Enhanced results using different modules. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
read the original abstract

Severe color casts, low contrast and blurriness of underwater images caused by light absorption and scattering result in a difficult task for exploring underwater environments. Different from most of previous underwater image enhancement methods that compute light attenuation along object-camera path through hazy image formation model, we propose a novel jointly wavelength compensation and dehazing network (JWCDN) that takes into account the wavelength attenuation along surface-object path and the scattering along object-camera path simultaneously. By embedding a simplified underwater formation model into generative adversarial network, we can jointly estimates the transmission map, wavelength attenuation and background light via different network modules, and uses the simplified underwater image formation model to recover degraded underwater images. Especially, a multi-scale densely connected encoder-decoder network is proposed to leverage features from multiple layers for estimating the transmission map. To further improve the recovered image, we use an edge preserving network module to enhance the detail of the recovered image. Moreover, to train the proposed network, we propose a novel underwater image synthesis method that generates underwater images with inherent optical properties of different water types. The synthesis method can simulate the color, contrast and blurriness appearance of real-world underwater environments simultaneously. Extensive experiments on synthetic and real-world underwater images demonstrate that the proposed method yields comparable or better results on both subjective and objective assessments, compared with several state-of-the-art 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 / 1 minor

Summary. The manuscript proposes a Jointly Adversarial Network (JWCDN) for underwater image enhancement that simultaneously performs wavelength compensation and dehazing by embedding a simplified underwater image formation model into a generative adversarial network. Key components include a multi-scale densely connected encoder-decoder for transmission map estimation and an edge-preserving module for detail recovery. A novel synthesis procedure is introduced to generate training images that incorporate the inherent optical properties of different water types, and the authors report that extensive experiments on synthetic and real underwater images show the method yields comparable or better subjective and objective results relative to several state-of-the-art approaches.

Significance. If the central claims hold, the work offers a physically motivated, end-to-end framework that jointly models surface-object attenuation and object-camera scattering, which could benefit marine vision tasks. The synthesis-based training strategy is a constructive element, but its unverified fidelity to real optical statistics limits the assessed significance.

major comments (2)
  1. [Method (synthesis subsection)] The novel underwater image synthesis method (described in the method section) is asserted to 'simulate the color, contrast and blurriness appearance of real-world underwater environments simultaneously' across water types, yet no quantitative validation—such as histogram distances, distributions of no-reference metrics on held-out real photographs, or statistical comparisons of synthetic versus real image statistics—is reported. This directly undermines the generalization claim from synthetic training data to real test images that supports the central performance assertion.
  2. [Experiments] The experiments section claims 'comparable or better results on both subjective and objective assessments' but supplies neither the concrete metric values (e.g., PSNR, SSIM, UCIQE), dataset cardinalities, nor any statistical significance testing or ablation results that would allow verification of the superiority claim over the cited baselines.
minor comments (1)
  1. [Method] Notation for the embedded formation model parameters (transmission, wavelength attenuation coefficients, background light) should be introduced with explicit equations early in the method section to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and outline revisions that will strengthen the presentation of our synthesis method and experimental results.

read point-by-point responses

  1. Referee: [Method (synthesis subsection)] The novel underwater image synthesis method (described in the method section) is asserted to 'simulate the color, contrast and blurriness appearance of real-world underwater environments simultaneously' across water types, yet no quantitative validation—such as histogram distances, distributions of no-reference metrics on held-out real photographs, or statistical comparisons of synthetic versus real image statistics—is reported. This directly undermines the generalization claim from synthetic training data to real test images that supports the central performance assertion.

    Authors: We agree that quantitative validation of the synthesis procedure would provide stronger support for its fidelity to real optical statistics. The current manuscript relies on qualitative visual similarity and downstream performance on real images to indicate realism, but we will add explicit comparisons in the revised synthesis subsection, including color histogram distances, distributions of no-reference metrics (e.g., UCIQE, BRISQUE) on held-out real photographs versus synthetic images, and basic statistical measures across water types. These additions will directly address the generalization concern. revision: yes

  2. Referee: [Experiments] The experiments section claims 'comparable or better results on both subjective and objective assessments' but supplies neither the concrete metric values (e.g., PSNR, SSIM, UCIQE), dataset cardinalities, nor any statistical significance testing or ablation results that would allow verification of the superiority claim over the cited baselines.

    Authors: We acknowledge that explicit numerical reporting, dataset sizes, ablations, and statistical tests would improve verifiability. While the manuscript includes objective evaluations via figures and some comparative tables, we will expand the experiments section in revision to include a summary table with concrete PSNR, SSIM, and UCIQE values, explicit dataset cardinalities for synthetic and real test sets, ablation results for the multi-scale encoder-decoder and edge-preserving modules, and basic statistical significance indicators (e.g., paired t-tests on metric differences) where appropriate. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's derivation embeds a simplified underwater image formation model into a GAN architecture (JWCDN) to jointly estimate transmission, wavelength attenuation, and background light, then recovers the image via the same model; a separate novel synthesis procedure generates training data with claimed optical properties. The central claim of comparable or superior performance rests on direct comparisons against external state-of-the-art methods on both synthetic and held-out real images. No quoted equation or step reduces a reported prediction, uniqueness result, or performance metric to a fitted parameter or self-citation by construction. The synthesis method and real-image evaluation are presented as independent of the network outputs, rendering the chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that the simplified formation model captures real degradation and that the synthesis procedure produces sufficiently realistic training data; no free parameters or invented physical entities are explicitly quantified in the abstract.

axioms (1)
  • domain assumption Simplified underwater image formation model accurately represents combined wavelength attenuation and scattering effects
    The method embeds this model into the GAN to recover images.
invented entities (1)
  • JWCDN with multi-scale dense encoder-decoder and edge-preserving modules no independent evidence
    purpose: Joint estimation of transmission map, wavelength attenuation, and background light
    New network architecture proposed for the joint task.

pith-pipeline@v0.9.0 · 5784 in / 1445 out tokens · 45007 ms · 2026-05-24T22:37:49.762545+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

47 extracted references · 47 canonical work pages · 2 internal anchors

  1. [1]

    Ludvigsen, B

    M. Ludvigsen, B. Sortland, G. Johnsen, and H. Singh, ”Applications of geo-referenced underwater photo mosaics in marine biology and archaeology,” Oceanography, vol. 20, no. 4, pp. 140-149, 2007

    work page 2007

  2. [2]

    Bonin, A

    F. Bonin, A. Burguera, and G. Oliver, ”Imaging systems for advanced underwater vehicles,” J. Mar. Res., vol. 8, no. 8, pp. 65-86, 2011

    work page 2011

  3. [3]

    Schettini and S

    R. Schettini and S. Corchs, ”Underwater Image Processing: State of the Art of Restoration and Image Enhancement Methods,” EURASIP J. Adv. Signal Process., vol. 2010, no. 1, pp. 746052, 2010

    work page 2010

  4. [4]

    Schechner and N

    Y . Schechner and N. Karpel, ”Clear underwater vision,” in Proc. IEEE CVPR, vol. 1, pp. 536543, Jun 2004

    work page 2004

  5. [5]

    Sheinin and Y

    M. Sheinin and Y . Schechner, ”The next best underwater view,” in Proc. IEEE CVPR, pp. 3764-3773, Jun, 2016

    work page 2016

  6. [6]

    McGlamery, ”A computer model for underwater camera systems,” in Proc

    B. McGlamery, ”A computer model for underwater camera systems,” in Proc. SPIE, Ocean Opt. VI , vol. 208, pp. 221-232, 1980

    work page 1980

  7. [7]

    Jaffe, ”Computer modeling and the design of optimal underwater imaging-systems,” IEEE J

    J. Jaffe, ”Computer modeling and the design of optimal underwater imaging-systems,” IEEE J. Ocean. Eng. , vol. 15, no. 2, pp. 101-111, 1990

    work page 1990

  8. [8]

    X. Zhao, T. Jin, and S. Qu, ”Deriving inherent optical properties from background color and underwater image enhancement,” Ocean Eng., vol. 94, pp. 163-172, 2015

    work page 2015

  9. [9]

    Chiang and Y

    J. Chiang and Y . Chen, ”Underwater image enhancement by wavelength compensation and dehazing,” IEEE Trans. Image Process. , vol. 21, no. 4, pp. 1756-1769, 2012

    work page 2012

  10. [10]

    C. Li, J. Guo, R. Cong, Y . Pang, and B. Wang, ”Underwater image enhancement by dehazing with minimum information loss and histogram distribution prior,” IEEE Trans. Image Process., vol. 25, no. 12, pp. 5664- 5677, 2016

    work page 2016

  11. [11]

    Drews, E

    P. Drews, E. Nascimento, S. Botelho and M. Campos, ”Underwater depth estimation and image restoration based on single images,” IEEE Comput. Graph. Appl., vol. 36, no. 2, pp. 24-35, 2016

    work page 2016

  12. [12]

    K. He, J. Sun, and X. Tang, ”Single image haze removal using dark channel prior,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 33, no. 12, pp. 2341-2343, 2011

    work page 2011

  13. [13]

    Y . Peng, K. Cao, and P. Cosman, ”Generalization of the dark channel prior for single image restoration,” IEEE Trans. Image Process., vol. 27, no. 6, pp. 2856-2868, 2018. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 13

    work page 2018

  14. [14]

    Galdran, D

    A. Galdran, D. Pardo, A. Picn, and A. Alvarez-Gila, ”Automatic red- channel underwater image restoration,”J. Vis. Commun. Image Represent., vol. 26, pp. 132-145, 2015

    work page 2015

  15. [15]

    Peng and P

    Y . Peng and P. Cosman, ”Underwater image restoration based on image blurriness and light absorption,” IEEE Trans. Image Process., vol. 26, no. 4, pp. 1579-1594, 2017

    work page 2017

  16. [16]

    Underwater Single Image Color Restoration Using Haze-Lines and a New Quantitative Dataset

    D. Berman, D. Levy, S. Avidan, and T. Treibitz, ”Underwater Single Image Color Restoration Using Haze-Lines and a New Quantitative Dataset,” arXiv preprint arXiv:1811.01343 , 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  17. [17]

    X. Fu, P. Zhuang, Y . Huang, Y . Liao, X. Zhang, and X. Ding, ”A retinex- based enhancing approach for single underwater image,” In Proc. IEEE Int. Conf. on Imag. Process. (ICIP) , pp. 4572-4576, Oct. 2014

    work page 2014

  18. [18]

    Ancuti and C

    C. Ancuti and C. O. Ancuti, ”Enhancing underwater images and videos by fusion,” in Proc. IEEE CVPR , pp. 81-88, Jun. 2012

    work page 2012

  19. [19]

    Ancuti, C

    C. Ancuti, C. O. Ancuti and C. Vleeschouwer, ”Color balance and fusion for underwater image enhancement,” IEEE Trans. Image Process., vol. 27, no. 1, pp. 379-393, 2018

    work page 2018

  20. [20]

    Ghani and A

    A. Ghani and A. IsaN, ”Underwater image quality enhancement through integrated color model with rayleigh distribution,” Appl. Soft Comput. , vol. 27, pp. 219-230, 2015

    work page 2015

  21. [21]

    Ghani and A

    A. Ghani and A. IsaN, ”Enhancement of low quality underwater im- age through integrated global and local contrast correction,” Appl. Soft Comput., vol. 37, pp. 332-344, 2015

    work page 2015

  22. [22]

    C. Li, J. Guo, and C. Guo, ”Emerging from water: underwater image color correction based on weakly supervised color transfer,” IEEE Signal Process. Lett., vol. 25, no. 3, pp. 323-327, 2018

    work page 2018

  23. [23]

    W. Ren, S. Liu, H. Zhang, J. Pan, X. Cao, and M.-H. Yang, ”Single image dehazing via multi-scale convolutional neural networks,” In ECCV, pp. 154-169, 2016

    work page 2016

  24. [24]

    B. Cai, X. Xu, K. Jia, C. Qing, and D. Tao, ”Dehazenet: An end-to-end system for single image haze removal,” IEEE Trans. Image Process., vol. 25, no. 11, pp. 5187-5198, 2016

    work page 2016

  25. [25]

    B. Li, X. Peng, Z. Wang, J. Xu, and D. Feng, ”An all-in-one network for dehazing and beyond,” in Proc. IEEE ICCV , 2017

    work page 2017

  26. [26]

    Zhang and V

    H. Zhang and V . Patel, ”Densely connected pyramid dehazing network,” Proc. IEEE CVPR , pp. 3194-3203, 2018

    work page 2018

  27. [27]

    J. Li, K. Skinner, R. Eustice, and M. Roberson, ”WaterGAN: un- supervised generative network to enable real-time color correction of monocular underwater images,” IEEE Robotics and Automation Letters , vol. 3, no. 1, pp. 387-394, 2017

    work page 2017

  28. [28]

    Fabbri, M

    C. Fabbri, M. Islam, and J. Sattar, ”Enhancing underwater imagery using generative adversarial networks,” IEEE International Conference on Robotics and Automation , pp. 7159-7165, 2018

    work page 2018

  29. [29]

    C. Li, J. Guo, C. Guo, R. Cong, and J. Gong, ”A hybrid method for underwater image correction,” Pattern Recognit. Lett., vol. 94, pp. 62-67, 2017

    work page 2017

  30. [30]

    Huang, Z

    G. Huang, Z. Liu, K. Q. Weinberger, L. van der Maaten. ”Densely connected convolutional networks,” in Proc. IEEE CVPR, pp. 4700-4708, 2017

    work page 2017

  31. [31]

    Ronneberger, P

    O. Ronneberger, P. Fischer, and T. Brox, ”U-net: Convolutional networks for biomedical image segmentation,” in International Conference on Medical Image Computing and Computer-Assisted Intervention , pp. 234- 241, 2015

    work page 2015

  32. [32]

    S. Jgou, M. Drozdzal, D. Vazquez, A. Romero, and Y . Bengio, ”The one hundred layers tiramisu: Fully convolutional densenets for semantic segmentation,” in Proc. IEEE CVPRW , pp. 11751183, 2017

    work page 2017

  33. [33]

    Zhu and S

    Y . Zhu and S. Newsam, ”Densenet for dense flow,” in Proc. IEEE Int. Conf. on Imag. Process. (ICIP) , 2017

    work page 2017

  34. [34]

    H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, ”Pyramid scene parsing network,” in Proc. IEEE CVPR , pp. 2881-2890, 2017

    work page 2017

  35. [35]

    Ioffe and C

    S. Ioffe and C. Szegedy, ”Batch normalization: Acceleratingdeep net- work training by reducing internal covariate shift,” in International Conference on Machine Learning , pp. 448-456, 2015

    work page 2015

  36. [36]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, ”Generative adversarial nets,” in Advances in Neural Information Processing Systems , pp. 2672-2680, 2014

    work page 2014

  37. [37]

    Isola, J

    P. Isola, J. Zhu, T. Zhou, and A. Efros, ”Image-to-image translation with conditional adversarial networks,” in Proc. IEEE CVPR , pp. 1125-1134, 2017

    work page 2017

  38. [38]

    Johnson, A

    J. Johnson, A. Alahi, and L. Fei-Fei, ”Perceptual losses for real-time style transfer and super-resolution,” in ECCV, pp. 694-711, 2016

    work page 2016

  39. [39]

    Ummenhofer, H

    B. Ummenhofer, H. Zhou, J. Uhrig, N. Mayer, E. Ilg, A. Dosovitskiy, and T. Brox, ”Demon: Depth and motion network for learning monocular stereo,” in Proc. IEEE CVPR , pp. 5038-5047, 2017

    work page 2017

  40. [40]

    Silberman, P

    N. Silberman, P. Kohli, D. Hoiem, and R. Fergus, ”Indoor segmentation and support inference from rgbd images,” in ECCV, pp. 746-760, 2012

    work page 2012

  41. [41]

    Deep Underwater Image Enhancement

    S. Anwar, C. Li, F. Porikli, ”Deep Underwater Image Enhancement,” arXiv preprint arXiv:1807.03528 , 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  42. [42]

    J. Zhu, T. Park, P. Isola, and A. Efros, ”Unpaired image-to-image translation using cycle-consistent adversarial networks,” in Proc. IEEE ICCV, pp. 2223-2232, 2017

    work page 2017

  43. [43]

    Jerlov, ”Irradiance optical classification,” in Optical Oceanography, pp

    N. Jerlov, ”Irradiance optical classification,” in Optical Oceanography, pp. 118120, 1968

    work page 1968

  44. [44]

    M. G. Solonenko and C. D. Mobley, ”Inherent optical properties of Jerlov water types,” Appl. Opt., vol. 54, no. 17, pp. 5392-5401, 2015

    work page 2015

  45. [45]

    S. Wang, K. Ma, H. Yeganeh, Z. Wang and W. Lin, ”A patch-structure representation method for quality assessment of contrast changed images.” IEEE Signal Process. Lett. , vol. 22, no. 1,pp. 2387-2390, 2015

    work page 2015

  46. [46]

    Crete, T

    F. Crete, T. Dolmiere, P. Ladret, and M. Nicolas, ”The Blur Effect: Perception and Estimation with a New No-Reference Perceptual Blur Metric.” in Human Vision and Electronic Imaging, vol. 6492, pp. 64920I, 2007

    work page 2007

  47. [47]

    Panetta, C

    K. Panetta, C. Gao, and S. Agaian, ”Human-visual-system-inspired underwater image quality measures.” IEEE J. Ocean. Eng. , vol. 41, no. 3, pp. 541-551, 2016

    work page 2016