Enhancement of Underwater Images by CNN-Based Color Balance and Dehazing
Round 1
Reviewer 1 Report
The paper presents an interesting idea but some parts should be modified:
- Even if in the introduction there are some references to the state of the art, an ad hoc section should be created to clarify the problem that is being faced;
- Figure 2 is important as it graphically displays the proposed framework. It should be better contextualized and commented (in the caption);
- A description of the dataset used is missing (perhaps in a table);
- Where are the comparison algorithms used with the proposed technique described, albeit briefly?
- In the specific context in which CNN is used to process images, the following paper should be cited:
Manzo, Mario, and Simone Pellino. "Voting in transfer learning system for ground-based cloud classification." Machine Learning and Knowledge Extraction 3, no. 3 (2021): 542-553.
Author Response
First of all, the authors sincerely appreciate the reviewer for his/her help with the manuscript. Below are point-by-point responses to the reviewer’s comments.
Comment 1: Even if in the introduction there are some references to the state of the art, an ad hoc section should be created to clarify the problem that is being faced;
Response: Thanks for the suggestion. In the revised manuscript, an ad hoc section, Literature review, has been added to report the state of the art and clarify the problem that is being faced.
Comment 2: Figure 2 is important as it graphically displays the proposed framework. It should be better contextualized and commented (in the caption);
Response: Thanks for the suggestion. In the revised manuscript, more explanation of figure 2 has been supplemented and the caption has been revised.
Comment 3: A description of the dataset used is missing (perhaps in a table)
Response: Thanks for the comment. In the revised manuscript, some general description of four used datasets has been added, in the first paragraph of Section 5.
Comment 4: Where are the comparison algorithms used with the proposed technique described, albeit briefly?
Response: Thanks for the comment. In the revised manuscript, the brief descriptions of the comparison algorithms are added, after the first paragraph of Section 5.
Comment 5: In the specific context in which CNN is used to process images, the following paper should be cited:
Manzo, Mario, and Simone Pellino. "Voting in transfer learning system for ground-based cloud classification." Machine Learning and Knowledge Extraction 3, no. 3 (2021): 542-553.
Response: Thanks for the suggestion. In the revised manuscript, this paper has been cited in the introduction section, and added to the list of references.
Reviewer 2 Report
Dear authors,
When someone looks through your manuscript, he/she cannot easily follow the idea of the paper. For example, it is not easy to find what is the proposal of the paper, what are the methods or algorithms proposed.
You should rename the titles in section 3, and reorganize this section.
In the introduction, you should explicitly state aims, goals and contributions of the paper.
Is the Section 2 necessary? It states known facts.
Figure 1 looks empty. You could make it looks full by adding text in squares.
Figure 2 - "t" is not visible in "output".
Line 160 is just "."?
Figure 6: try to thicken the blue lines.
Used fusion algorithm is not elaborated sufficiently.
Kind regards
Author Response
First of all, the authors sincerely appreciate the reviewer for his/her help with the manuscript. Below are point-by-point responses to the reviewer’s comments.
Comment 1: You should rename the titles in section 3, and reorganize this section.
Response: Thanks for the suggestion. In the revised manuscript, the titles in section 4 (former section 3) have been renamed, including the titles of section 4, subsection 4.2 and subsection 4.3. Moreover, some explanations have been added in this section, including the explanation of figure 2, the explanation of figure 4 and the explanation of the adaptive contrast enhancement algorithm.
Comment 2: In the introduction, you should explicitly state aims, goals and contributions of the paper.
Response: Thanks for the comment. In the revision, the goal of the study has been explicitly explained in the second paragraph of introduction section. The contributions of the paper have also been explicitly stated, at the end of a newly added section titled with Literature review.
Comment 3: Is the Section 2 necessary? It states known facts.
Response: Thanks for the comment. In the revision, section 2 has been updated as section 3. Indeed, this section states known facts. Nevertheless, the authors think the fundamental knowledge on the underwater imaging and CNN is valuable and helpful for those beginners in the field of underwater image processing.
Comment 4: Figure 1 looks empty. You could make it looks full by adding text in squares.
Response: Thanks for your suggestion. In the revised manuscript, texts have been added in squares in figure 1. Corresponding explanation has also been added in the paragraph above the figure.
Comment 5: Figure 2 - "t" is not visible in "output".
Response: Thanks for pointing out the mistake. In the revised manuscript, the textbox has been moved to a proper position to make “output” visible completely.
Comment 6: Line 160 is just "."?
Response: Thanks for pointing out the mistake. In the revised manuscript, the word “of ” has been removed from the sentence so that the period can be displayed correctly.
Comment 7: Figure 6: try to thicken the blue lines.
Response: Thanks for the comment. In the revised manuscript, the blue lines has been thickened.
Comment 8: Used fusion algorithm is not elaborated sufficiently.
Response: Thanks for the comment. In the revised manuscript, the explanation of fusion algorithm has been added before figure 7. Moreover, the expression of fusion algorithm was given as Eq.(26).
Reviewer 3 Report
Authors contribution:
The problem that the authors have defined is that the particularity of underwater environment degrades the visibility and quality of underwater images and videos for example resultant color deviation, blur and decrease of contrast. The attenuation and scattering of light propagation in water partly account for the degradation.
To improve the clarity of underwater images, the authors have proposed an all-in-one dehazing model which aims to reduce the accumulated errors that result from determining the transmittance and background light individually when a conventional restoration model is used.
The enhancement strategy, that the authors have proposed involves three modules:
1.A CNN based generative adversarial network (GAN) is constructed to achieve color balance of underwater images.
2.The conventional imaging model is modified to an all-in-one model in which a comprehensive index is introduced and estimated by a deep CNN
3.By using a fusion algorithm, the overall contrast and increase local details are improved.
To demonstrate the effectiveness of the authors method, five algorithms are compared. Comparison is conducted from subjective visual effects and objective evaluation.
I have some reviewer notes:
Abstract. Line 18. Better than what…? It will be good to add one sentence at the end of the abstract, how your work will be continued. Also, what is the accuracy of the proposed algorithms?
Equation 3. You have to describe all of the variables that are in the formula, not only “g(x)”. Also, for equations (4), (5), (8), (9).
Figure 1. It is not good to present empty boxes. Write in the boxes what are they about.
Figure 4. Show axis titles both for horizontal and vertical axes.
Equation 19. you have to cite literature source for this grey level calculation.
Figure 8. If it is possible, write in the description of the figure, where are these images from. For example, with WGS84 coordinates. Also, for Figure 9.
Equation 29. You have to describe if the result is dimensionless.
Tables 1-4. What are the measurement units of the presented values? You have to describe it.
Discussion part is missing. You have to compare your results with minimum 3 papers from other authors.
Conclusion part. It is not clear how your work improve the known solutions in this study area. Also, you have to describe what is the accuracy of your method. With values, if it is possible. Do not cite figures in the Conclusion part. Make summarization of the results, and also how it improves the known solutions.
I have some suggestions:
Make better structure of your equations. Describe the used variables in appropriate way. Make more comparative analyses with results from other authors. Present the accuracy of your results with values. These suggestions will improve your contribution.
Author Response
First of all, the authors sincerely appreciate the reviewer for his/her help with the manuscript. Below are point-by-point responses to the reviewer’s comments.
Comment 1: Abstract. Line 18. Better than what…? It will be good to add one sentence at the end of the abstract, how your work will be continued. Also, what is the accuracy of the proposed algorithms?
Response: Thanks for the comments. In the revised manuscript, the sentence in line 18 has been revised as “The comparison results indicate that the proposed method gains better enhancement effects for underwater images in different scenes than the other enhancement algorithms,” The advantages of the proposed algorithms have also been added in the abstract. At the end of the paper, a statement has been added to explain how the work will be continued.
Comment 2: Equation 3. You have to describe all of the variables that are in the formula, not only “g(x)”. Also, for equations (4), (5), (8), (9).
Response: Thanks for the comments. In Eq.(3), except for g(x), the other two variables, J(x) and t(x), have been explained in the former equation, i.e. Eq.(2). So did equations (4) and (5). In the revision, the explanations of D(x) and G(z) in Eq.(8) have been added. For Eq.(9), two variables were explained in the following equations (10) and (11), respectively.
Comment 3: Figure 1. It is not good to present empty boxes. Write in the boxes what are they about.
Response: Thanks for your suggestion. In the revised manuscript, texts have been added in squares in figure 1. Corresponding explanation has also been added in the paragraph above the figure.
Comment 4: Figure 4. Show axis titles both for horizontal and vertical axes.
Response: Thanks for the comment. Figure 4 is a network structure diagram, not a result image instead. Therefore, no titles were given for horizontal and vertical axes.
Comment 5: Equation 19. you have to cite literature source for this grey level calculation.
Response: Thanks for your suggestion. In the revised manuscript, the related reference has been added for this equation.
Comment 6: Figure 8. If it is possible, write in the description of the figure, where are these images from. For example, with WGS84 coordinates. Also, for Figure 9.
Response: Thanks for the comment. These images are taken from free datasets. As understood by the authors, there are no WGS84 coordinates involved in these images.
Comment 7: Equation 29. You have to describe if the result is dimensionless.
Response: Thanks for the comment. In the revised manuscript, it has been mentioned that the four metrics (including the referred equation) are dimensionless, in the first paragraph of subsection 5.3.
Comment 8: Tables 1-4. What are the measurement units of the presented values? You have to describe it.
Response: Thanks for the comment. As replied above, the results calculated by equations (34)~(37) are dimensionless. Therefore, no units are defined for the values in Tables 1-4 .
Comment 9: Discussion part is missing. You have to compare your results with minimum 3 papers from other authors.
Response: Thanks for the comment. The proposed algorithm are compared with other five algorithms (including MSRCR, RCP, UDCP, ICM, and RGHS). In the revised manuscript, the five algorithms are briefly described and the corresponding references (papers) have been added, before subsection 5.3.
Comment 10: Conclusion part. It is not clear how your work improve the known solutions in this study area. Also, you have to describe what is the accuracy of your method. With values, if it is possible. Do not cite figures in the Conclusion part. Make summarization of the results, and also how it improves the known solutions.
Response: Thanks for the comment. Below figure 8, the advantages of the proposed method over known solutions in the underwater image enhancement are illustrated. The goal of the study is to enhance underwater images, rather than detect or recognize underwater object, therefore the accuracy of the proposed method might be understood from the evaluation metrics listed in tables 1-4. In the revision, the citation of figure has been removed in the conclusion section. Moreover, some sentences have been added to explain the results and how the proposed method improves the known solutions, at the end of the first paragraph of conclusion section.
Concluded comment: Make better structure of your equations. Describe the used variables in appropriate way. Make more comparative analyses with results from other authors. Present the accuracy of your results with values. These suggestions will improve your contribution.
Response: Thanks for the kind suggestion. In revising the manuscript, the authors check the equations carefully to express them correctly. Moreover, more discussion on the results compared with other authors has been supplemented. The authors sincerely appreciate the reviewer for his/her help to improve the paper.
Reviewer 4 Report
The paper proposes the Generative Adversarial Network (GAN) based method for underwater image dehazing. The paper needs to be revised and improved to address the comments and suggestions presented in the comments below.
1. Add the summary of results in the abstract section.
2. Explicitly state the novelty of this study and its contribution to the research field.
3. The overview of related works in the introduction section lacks of structure. You can use the existing overview papers such as for example, * Singh, et al. (2022). Visibility enhancement and dehazing: Research contribution challenges and direction. * Arif, et al. (2022). Comprehensive review of machine learning (ML) in image defogging: Taxonomy of concepts, scenes, feature extraction, and classification techniques; as a starting point of such discussion.
4. Figure 4 should be explained in more detail. What is the meaning of box width and height? Each box corresponds to a layer? Add a legend to explain the meaning of different colors.
5. There are many different models how to convert RGB to grayscale. Provide a supporting reference for Eq. (19).
6. Line 355: provide a missing reference number.
7. Use boxplots to compare the results from Tables 1-4 visually.
8. Can you evaluate the quality of edge detection quantitatively?
9. Discuss the limitations of the proposed methodology.
10. Improve the conclusions. Support your claims with the main numerical results from experiments.
Author Response
First of all, the authors sincerely appreciate the reviewer for his/her help with the manuscript. Below are point-by-point responses to the reviewer’s comments.
Comment 1: Add the summary of results in the abstract section.
Response: Thanks for the suggestion. In the revised manuscript, the summary of results has been added, at the end of abstract.
Comment 2: Explicitly state the novelty of this study and its contribution to the research field.
Response: Thanks for the comment. In the revision, the novelty of the study and the contributions have been explicitly stated, at the end of section 2.
Comment 3: The overview of related works in the introduction section lacks of structure. You can use the existing overview papers such as for example, * Singh, et al. (2022). Visibility enhancement and dehazing: Research contribution challenges and direction. * Arif, et al. (2022). Comprehensive review of machine learning (ML) in image defogging: Taxonomy of concepts, scenes, feature extraction, and classification techniques; as a starting point of such discussion.
Response: Thank for the advice. In the revision, these two papers (Ref.[35] and [36]) have been added as a starting point of the discussion on the application of deep learning to image enhancement, in the third paragraph of section 2.
Comment 4: Figure 4 should be explained in more detail. What is the meaning of box width and height? Each box corresponds to a layer? Add a legend to explain the meaning of different colors.
Response: Thanks for the suggestion. The width and height of boxes represent the width and height of feature map, and different boxes correspond to different convolutional layers of the network. In the revision, explanation of the boxes has been added. A legend has also been added in the figure explain the meaning of different colors.
Comment 5: There are many different models how to convert RGB to grayscale. Provide a supporting reference for Eq. (19).
Response: Thanks for your suggestion. In the revised manuscript, the related reference has been added for this equation.
Comment 6: Line 355: provide a missing reference number.
Response: Thanks for your suggestion. In the revised manuscript, the reference number has been provided.
Comment 7: Use boxplots to compare the results from Tables 1-4 visually.
Response: Thanks for your suggestion. In the revised manuscript, four figures (Figures 10-13) are added in which boxplots are used to compare the results from Tables 1-4 visually.
Comment 8: Can you evaluate the quality of edge detection quantitatively?
Response: Thanks for the question. From the comparison results of edge detection, it can be seen that the proposed method obviously gains more edge information in images than the other algorithms. Therefore, the study did not further perform the comparison quantitatively. Frankly, it is difficult for the authors to finish the required evaluation in a short time. Anyhow, the authors appreciate the reviewer’s suggestion and will conduct the recommended research in the future.
Comment 9: Discuss the limitations of the proposed methodology.
Response: Thanks for the suggestion. In the revised manuscript, the limitations of the proposed methodology has been supplemented, at the end of conclusion section.
Comment 10: Improve the conclusions. Support your claims with the main numerical results from experiments.
Response: Thanks for the comment. Some sentences have been added to explain the results and how the proposed method improves the known solutions, at the end of the first paragraph of conclusion section.
Round 2
Reviewer 1 Report
as far as I'm concerned, the paper can be accepted in this form
Author Response
The authors sincerely appreciate the reviewer for his/her help with the manuscript.
Reviewer 2 Report
Thanks for answers. Corrections are mostly fine. However, it seems that your system requires high computational costs. What are the requirements for the proposed to function online? Can you provide the execution speed data?
Author Response
The authors sincerely appreciate the reviewer for his/her help with the manuscript. Below are the responses to the reviewer's comments.
Comment: However, it seems that your system requires high computational costs. What are the requirements for the proposed to function online? Can you provide the execution speed data?
Response: Thanks for the comments. The method proposed in this paper is a combination of three algorithms, therefore it is time-consuming. At present, it is not applied to the online enhancement. It is believed that the computational cost would reduce with the upgrade of computer and the improvement of algorithm, as stated in the conclusion. In the revised manuscript, the running environment in computer and the execution speeds for different algorithms have been given in the added Table 1.
Reviewer 4 Report
The paper was well revised, while all my comments were addressed appropriately. The manuscript is suitable for publication. The authors, however, should carefully check the language for errors and typos such as "Entorpy" (p. 17).
Author Response
The authors sincerely appreciate the reviewer for his/her help with the manuscript. Below is the response to the reviewer's comment.
Comment: The authors, however, should carefully check the language for errors and typos such as "Entorpy" (p. 17).
Response: Thanks for the comment and pointing out the mistake. In the revision, "Entorpy" (p.17) has been revised as "Entropy". Moreover, the authors check the submission carefully to correct language errors and typos. All changes have been highlighted in the revision.
