Single Remote Sensing Image Dehazing Using Robust Light-Dark Prior
Round 1
Reviewer 1 Report
Single Remote Sensing Image De-hazing Using Robust Light-dark prior
General description:
In the present work, a single remote sensing image de-hazing method based on robust light-dark prior (RLDP) is developed, which utilizes a hybrid model and is robust to outlier pixels. The RLDP method, firstly removes the haze by a robust dark channel prior (RDCP), and then removes the shadow with a robust light channel prior (RLCP). Furthermore, a cube RME (CRME)-based stable state search criterion is applied to solve the problem of patch size setting. Experimental results well illustrated demonstrate that the RLDP method can effectively remove the haze from optical images.
Remarks:
Page 1, row 7: Disclose the abbreviation RME.
Page 4, row 110: It is written: “haze-free but shadowyg image”. Correct: haze-free but shadowing image.
Page 4, row 111. It is recommended to write d(x) in the exponent.
It is recommended to read carefully the text and correct if necessary.
Comments for author File: 
Comments.pdf
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript.
Comment 1. Page 1, row 7: Disclose the abbreviation RME.
【Response】: Thank you for this comment. We have disclosed the abbreviation RME as root mean enhancement.
Revised version: Further, a cube root mean enhancement (CRME)-based stable state search criterion is proposed for solving the difficult problem of patch size setting.
Comment 2. Page 4, row 110: It is written: “haze-free but shadowyg image”. Correct: haze-free but shadowing image.
【Response】: Thank you for your careful review. We have corrected the word and checked for similar issues throughout the manuscript.
Revised version: …haze-free but shadowing image.
Comment 3. Page 4, row 111. It is recommended to write d(x) in the exponent.
【Response】: Thank you for the suggestion. We have modified d(x) at row 112 to d(x + γ) to make it consistent.
Revised version: …and s(x) = e−βd(x+γ) is the shadow transmission map, with β, d(x + γ) and γ being the atmospheric scattering parameter, the cloud transparency and bias respectively.
Comment 4. It is recommended to read carefully the text and correct if necessary.
【Response】: Thank you for this comment. We have made a major improvement to the manuscript and proofread it many times, and we hope that will make it more acceptable for publication.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments for author File: Comments.pdf
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript.
Comment 1. Why the RDCP and RLCP can make the algorithm robust? The authors should present more analysis.
【Response】: Thank you very much for your advice. We have added a description of the motivation for RDCP and RLCP in this revision.
Revised version: …. and improve the robustness of the algorithm, we set the maximum value of the patch to µ + 3σ.
Comment 2. For Eq.(2) and Eq.(3), the ts(x) is suggest to be t s(x).
【Response】: Thank you for the suggestion. We have modified ts(x) to s(x) in the whole manuscript.
Comment 3. Below Eq.(14), what is the parameter ‘minmaxv’?
【Response】: Thank you for your careful review. We have added a parameter description according to your prompt.
Revised version: …we take the minimum of the maximum of three channels (minmaxv).
Comment 4. In Eq.(16), what does the theta set to be ?
【Response】: Thank you for your helpful comment. We have added a description of this parameter.
Revised version: where θ is a tiny constant (fixed to 0.001 in this paper) …
Comment 5. The experimental results are insufficient, only Fig.3 to Fig.5 are presented. More results are suggested to be added and discussed.
【Response】: Thank you for this comment. We have added more image quality evaluation and negative aspects analysis in Qualitative Evaluations.
Revised version: Despite such advantages, the inhomogeneous fog removal effect still needs to be improved due to the bias in the parameter estimation, especially in the atmospheric light estimation (as shown in the 5th and 8th images).
Comment 6. Some classic references for model-based methods needs to be discussed. For instance: ’Fast image dehazing method based on linear transformation, 2017’,’ A Novel Fast Single Image Dehazing Algorithm Based on Artificial Multi-exposure Image Fusion, 2021’.
【Response】: Thank you for this comment. We have carefully studied these high-quality papers and revised the introduction.
Revised version: Wang et al. [12] proposed a fast algorithm to solve the linear assumption model based on dark channel. Zhu et al. [13] offered a pixel-wise fusion weight map to guide multi-exposure image fusion, which improves the performance and robustness of dehazing.
Comment 7. English expression needs improvement.
【Response】: We are very sorry for the inconvenience caused to your reading. We have revised the whole manuscript with the assistance of a colleague whose English is good, so we hope it can meet the journal’s standard.
Author Response File: Author Response.pdf
Reviewer 3 Report
- The paper proposes the improved Robust light dark prior (RLDP) that is robust to outlier pixels.
- Paper organization is good enough for publication.
- The survey is well organized and provides deep insights.
- The contribution is clearly described. The outlier pixels often cause the dark shadow. The proposed method solves this issue and is especially good for cloud images
- In Section 2, RLDP is well explained with graphical representations. The mathematical description is well organized and professional.
- In transmission estimation, the statistical approach is novel and quite reasonable to be robust to outlier pixels. Fig 2. well demonstrates the effect of the proposed statistical method.
- The experiments are conducted in deep. It pay that authors benchmarked the various existing methods. With figure 4,5 in the experiment, the proposed method can be verified to be good for cloud images.
- In qualitative evaluations, a reviewer suggests that the human subjective test should be conducted, for example, MOS test that is often used for evaluating image quality.
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript.
Comment 1. The paper proposes the improved Robust light dark prior (RLDP) that is robust to outlier pixels.
【Response】: Thank you for your valuable comments on this paper.
Comment 2. Paper organization is good enough for publication.
【Response】: Thank you for your valuable comments on this paper.
Comment 3. The survey is well organized and provides deep insights.
【Response】: Thank you for your valuable comments on this paper.
Comment 4. The contribution is clearly described. The outlier pixels often cause the dark shadow. The proposed method solves this issue and is especially good for cloud images.
【Response】: Thank you for your valuable comments on this paper.
Comment 5. In Section 2, RLDP is well explained with graphical representations. The mathematical description is well organized and professional.
【Response】: Thank you for your valuable comments on this paper.
Comment 6. In transmission estimation, the statistical approach is novel and quite reasonable to be robust to outlier pixels. Fig 2. well demonstrates the effect of the proposed statistical method.
【Response】: Thank you for your valuable comments on this paper.
Comment 7. The experiments are conducted in deep. It pay that authors benchmarked the various existing methods. With figure 4,5 in the experiment, the proposed method can be verified to be good for cloud images.
【Response】: Thank you for your valuable comments on this paper.
Comment 8. In qualitative evaluations, a reviewer suggests that the human subjective test should be conducted, for example, MOS test that is often used for evaluating image quality.
【Response】: Thank you for the suggestion. We have added an image quality evaluation index CIEDE2000, which is more consistent with subjective visual perception in the experimental part.
Revised version: Compared with the above two indicators, CIEDE2000 is more in line with subjective visual perception, and the lower the value, the smaller the color difference.
Author Response File: Author Response.pdf
Reviewer 4 Report
Please check attachment. Thank you
Comments for author File: Comments.pdf
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript.
Comment 1. The motivation is not clear. Please specify the importance of the proposed solution.
【Response】: Thank you for this comment. To better express the motivation of this manuscript, we first highlighted several problems of current prior-based methods at the end of the second paragraph of the Introduction, and then explained that the proposed method aims to solve these problems at the beginning of the fourth paragraph.
Revised version: Such methods are simple and easy to implement, but still suffer with the problems of easy distortion of image color, aggravation of shadow degree and difficulty in parameter setting.
Comment 2. The listed contributions are a little bit weak. Please highlight the innovations of the proposed solution.
【Response】: Thank you for the suggestion. We have added additional statements in the highlights section to better reflect the contributions of this manuscript.
Revised version: Then, a two-stage dehazing algorithm was proposed based on the haze hybrid model. …, which can adaptively adjust the patch size according to the contrast measure of a single image.
Comment 3. Authors ignore some recently published solutions, such as "A novel fast single image dehazing algorithm based on artificial multi-exposure image fusion", IEEE Transactions on Instrumentation and Measurement 70, 1-23, 2021 and Single-Image Dehazing Based on Improved Bright Channel Prior and Dark Channel Prior. Electronics 2023, 12, 299. Please discuss them.
【Response】: Thank you for this comment. We have carefully studied these high-quality papers and revised the introduction.
Revised version: Zhu et al. [13] offered a pixel-wise fusion weight map to guide multi-exposure image fusion, which improves the performance and robustness of dehazing. Li et al. [15] used bright and dark channel priors for dehazing the segmented sky and non-sky regions separately, and proposed a weighting method to fuse the parameters.
Comment 4. Please discuss how to obtain the suitable parameter values used in the proposed solution.
【Response】: Thank you for your helpful comment. We have added a description of this parameter setting.
Revised version: Appropriate patch size (sps) is defined as the search step (where patch size is monotonically increased from 20 to 120) when CRME reaches a stable state for the first time.
Generally, the larger θ value, the easier it is to reach a stable state, corresponding to a smaller patch size, and vice versa, corresponding to a larger patch size.
Comment 5. What is the experimental environment?
【Response】: Thank you for this comment. We have supplemented the experimental environment details in the Experimental Setup section.
Revised version: All experiments were performed on an Intel(R) Core (TM) i7-11700 @ 2.50 GHz and 32.0 GB RAM hardware environment with Python 3.9.
Comment 6. The experimental results are not convincing. Please compare the proposed solution with more recently published solution, especially the solutions published in 2022
【Response】: Thank you for this comment. We have added time tags for comparison algorithms in the Experimental Setup section, including 4 classical algorithms and 3 algorithms in the last two years.
Revised version: Eventually, we compared the dehazing effect of RLDP with DCP 2010[5], CL 2014[35], CEP 2017[9], HL 2018 [10], CEEF-TMM 2021[40], UNTV 2021[41], and ACT 2022[42].
Comment 7. More objective evaluation indicators should be introduced.
【Response】: Thank you for the suggestion. We have added an image quality evaluation index CIEDE2000, which is more consistent with subjective visual perception in the experimental part.
Revised version: Compared with the above two indicators, CIEDE2000 is more in line with subjective visual perception, and the lower the value, the smaller the color difference.
Comment 8. Please apply the proposed solution to different datasets.
【Response】: Thank you very much for your advice. We have added several real-world Landsat 8 RSI images as test datasets in the qualitative evaluation to better demonstrate the effectiveness of the proposed method.
Comment 9. "Robust" is mentioned many times in this manuscript. More explanations of "robust" should be given.
【Response】: Thank you very much for your advice. We have added a description of the motivation for robust of the manuscript.
Revised version: ….and improve the robustness of the algorithm, we set the maximum value of the patch to µ + 3σ.
Comment 10. More technical details of how to find an appropriate patch size should be given.
【Response】: Thank you for your helpful comment. We have added a description of the appropriate patch size searching details.
Revised version: Appropriate patch size (sps) is defined as the search step (where patch size is monotonically increased from 20 to 120) when CRME reaches a stable state for the first time.
Author Response File: Author Response.pdf
Reviewer 5 Report
The paper presents an interesting idea but should be modified in some of its parts:
- Although references have been made in the introduction but a state of the art section should be organised;
- Algorithm 1 is central to the presented paper. It should be better referenced in the text with specific references;
- Details of the datasets chosen for testing are missing. A table should be inserted which summarizes the main characteristics;
- Experimentation conducted should be better commented highlighting positive/negative aspects;
- What happens if you use an approach with features from different sources? A recent paper dealing with this aspect should be cited:
Maddalena, L., Granata, I., Giordano, M., Manzo, M., & Guarracino, M. R. (2022). Classifying Alzheimer's Disease using MRIs and Transcriptomic Data. In BIOIMAGING (pp. 70-79).
Author Response
We really appreciate you for your carefulness and conscientiousness. Your suggestions are really valuable and helpful for revising and improving our paper. According to your suggestions, we have made the following revisions on this manuscript.
Comment 1. Although references have been made in the introduction but a state of the art section should be organized.
【Response】: Thank you very much for your advice. We have reorganized the relevant content in the Introduction section into the new State of the Art section.
Comment 2. Algorithm 1 is central to the presented paper. It should be better referenced in the text with specific references.
【Response】: Thank you for the suggestion. We have added two descriptions to the algorithm references.
Revised version: This algorithm first assesses the global atmospheric light A by Equation (10), and then estimates the ti and si of each pixel i by Equation (12) and Equation (14). Finally, the image dehazing is completed through two stages: recovering the haze-free image J and restoring shadow-free image I.
Comment 3. Details of the datasets chosen for testing are missing. A table should be inserted which summarizes the main characteristics.
【Response】: Thank you for the suggestion. We have added a detailed description of the datasets in Experimental Setup.
Revised version: which included 108 Alpine images, 118 Flat images, 110 Sandy images, 54 Mountainous images, and 110 Sea images.
Comment 4. Experimentation conducted should be better commented highlighting positive/negative aspects
【Response】: Thank you for your helpful comment. We have added negative aspects analysis in Qualitative Evaluations.
Revised version: Despite such advantages, the inhomogeneous fog removal effect still needs to be improved due to the bias in the parameter estimation, especially in the atmospheric light estimation (as shown in the 5th and 8th images).
Comment 5. What happens if you use an approach with features from different sources? A recent paper dealing with this aspect should be cited: Maddalena, L., Granata, I., Giordano, M., Manzo, M., & Guarracino, M. R. (2022). Classifying Alzheimer's Disease using MRIs and Transcriptomic Data. In BIOIMAGING (pp. 70-79).
【Response】: Thank you for your very fascinating and meaningful comment. We had tested some dehazing algorithms for other types of images (such as underwater images, fundus images, and dust images) for RSI dehazing, and chose an underwater image dehazing method as one of the comparison algorithm UNTV 2021 [44]. Through a lot of experiments, we found that no method can win in all scenarios, and additional parameter adjustment is required to achieve better results.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report
All my concerns have been addressed. I recommend this paper for publication.
Reviewer 5 Report
The citation about comment 5 is missing:
Maddalena, L., Granata, I., Giordano, M., Manzo, M., & Guarracino, M. R. (2022). Classifying Alzheimer's Disease using MRIs and Transcriptomic Data. In BIOIMAGING (pp. 70-79).
