Blind Infrared Remote-Sensing Image Deblurring Algorithm via Edge Composite-Gradient Feature Prior and Detail Maintenance

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper proposes a new algorithm for deblurring infrared images. The introduction presents the problem and the common challenges. Section 2 discusses the related work. It sometimes cites several papers without even a brief explanation of what are they about (e.g., lines 68, 96, 169). But the paper is long enough as it is. I like the summary of contributions at the end of Section 2.

The proposed method uses a composite gradient feature prior, followed by an adaptive regularisation. The algorithm is expressed in mathematical equations with proofs. A flowchart gives a nice overview of the method. Figure 2 is not mentioned anywhere in the text. I think it would fit on line 331. The citation on line 490 uses a different font. Variables I and q on line 494 should be bold. The proposed algorithm is nicely summarised in pseudo-code at the end of Section 4.

Section 5 presents the experiments. The authors used images from 3 publicly available datasets. They applied an artificial blur to the images so that the deblurred result can be compared to the original. Three metrics were used to evaluate the quality of the results, including visual information fidelity which mimics human perception. The results are compared to 5 competitive algorithms, 2 older (2016, 2017) and 3 modern (2020 and newer). The proposed method performs better in the vast majority of the experiments. It would be interesting to see the computation times too.

The authors further show that the composite gradient feature alone does not perform as well as the complete proposed method including the adaptive regularisation. Further experiments show that the method is not sensitive to the parameter settings, namely the patch size and the regularisation weights, as long as they are in reasonable limits.

The paper is written in perfect English without typos. The text is well structured and the proposed algorithm is clearly explained. The experiments and the further discussion are thorough and they prove the effectiveness of the proposed method.

Author Response

We would like to express our sincere gratitude to the reviewer for the detailed feedback and constructive suggestions, which have significantly improved the quality and clarity of our manuscript. We have carefully considered all the comments and made the necessary revisions to address the concerns raised. Below are our responses to the specific points highlighted. 1. On the issue of not providing detailed explanations for some cited references in the related work section: Some references are repeatedly mentioned in the related work section, such as several papers in citations 41-47, which are cited multiple times in the context. Considering the similarity or commonality of these algorithms, we have provided a separate and detailed introduction to these methods later in the paper, and a summary of their key points is presented in line 168. Therefore, we chose not to provide repeated explanations of each individual paper in this section to avoid redundancy. Additionally, some references list multiple papers simultaneously to highlight the commonality of a certain problem or phenomenon, which is addressed in several papers. For example, in lines 70 and 98, we cite multiple related works to showcase the prevalence of the issue, providing a more comprehensive view of the research background and current state in the field. 2. On Figure 2 not being referenced: Thank you for your valuable suggestion. It is true that Figure 2 was not cited in the main text, now we have referenced it in line 30. We will double-check to ensure that Figure 2 is appropriately referenced throughout the manuscript and make any necessary adjustments. 3. On the font issue in line 490: We have carefully reviewed the manuscript and made the necessary font adjustments to ensure consistency throughout. Additionally, we have bolded the variables "I" and "q" in line 494 as requested. 4. On the discussion of computation times: Thank you for your insightful comment. Among all the TV regularization-based comparison algorithms, the PMP algorithm, due to its focus on simplifying the process, has the shortest computation time. While our method has a slightly more complex computation process, its calculation time is comparable to that of mainstream algorithms, thanks to the sparsity of the prior matrix and the optimized iterative process. We will further optimize the efficiency of the algorithm in future work and try to provide a more detailed comparison of computation times.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this paper, a blind deblurring model based on ded MAP framework is proposed, which uses the composite gradient feature (CGF) change of the blurred edge region. The former item shows a high degree of sparsity, which helps mitigate the pathology of blind images and eliminates the blurriness of counter-problems. When the fuzzy kernel is determined, the effective edge information pixels detected from the fuzzy image are allocated to the smooth intermediate hidden image to further enhance the structure of the intermediate hidden image and improve the precision of the fuzzy kernel. The variational model proposed in this paper is effectively implemented mathematically by alternate iteration technique, and verified by a large number of numerical experiments and comparison with the latest methods.

The manuscript is highly innovative, but the following questions should be considered before acceptance.

1.Is Figure 3 a schematic diagram in previous literature or was it drawn by the author of this manuscript? It's very vague and I can't understand the flow of this mechanism.

2.In the comparison of Figure 5, although the author has marked the effect before and after compensation in red, it is still not obvious enough for me to compare the obvious effect, so I suggest redrawing the picture or processing the experimental data.

3.Whether the numerical experimental results in FIG. 6-11 can be drawn in other forms? All of them are drawn in gray scale, which limits my understanding of the results.

4.The table is not a three-wire table.

5.Is Figure 12 drawn directly from the calculation software? Why so vague? 14 is the same.

6.

 

Comments on the Quality of English Language

Minior revision

Author Response

We would like to express our sincere gratitude to the reviewer for the detailed feedback and constructive suggestions, which have significantly improved the quality and clarity of our manuscript. We have carefully considered all the comments and made the necessary revisions to address the concerns raised. Below are our responses to the specific points highlighted.

1. Regarding the question about Figure 3: Figure 3 is a schematic diagram drawn by the authors based on the research content. In response to the reviewer's comment regarding the unclear nature of the diagram, we have redrawn it to ensure that the mechanism flow is clearer and more understandable.

2. Regarding the comparison in Figure 5 and the unclear effect: Thank you for your feedback. To address the issue of unclear comparison in Figure 5, we have reprocessed the comparison results through experiments and optimized the images to more clearly highlight the differences before and after compensation.

3. Regarding the numerical experimental results in Figures 6-11: For the numerical experimental results in Figures 6-11, we have applied local zooming to each image to present the results more intuitively and enhance readability. Additionally, we have presented the evaluation metrics of the experimental results in the form of bar charts to facilitate the comparison of the strengths and weaknesses of different methods.

4. Regarding the table format: Following the reviewer's suggestion, we have replaced the original table of comparison results with bar charts, which provide a more intuitive display of the relative performance of different methods.

5. Regarding the blurriness in Figures 12 and 14: To address the blurriness in Figures 12 and 14, we have redrawn these figures to ensure clearer images that better showcase the experimental results.

 

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper proposes a blind deblurring algorithm for infrared remote sensing images. The author gives a lot of formula derivations in Sections 3 and 4. These formula derivations seem reasonable, and the experimental results given in Section 5 also seem credible. However, there are two questionable points in this paper:

1) Although the author claims that infrared remote sensing images will be blurred by the atmosphere, the influence of the atmosphere on the image is generally corrected by atmospheric correction methods in the remote sensing field. Moreover, the blurring of infrared images is the result of the interaction between temperature and sensor. It is normal for infrared images to be blurred, which is completely different from the blurring caused by focusing and low resolution. Therefore, the paper studies the blurring problem of infrared remote sensing images, and the significance of this work is questionable; if the deblurring problem of infrared remote sensing images is effective or necessary, the author should conduct research from the perspective of solving practical application problems, rather than simply comparing or evaluating the results with indicators;

2) The results given in the experimental part of the paper cannot show that the processing of infrared remote sensing images by this method is truly effective based on visual inspection in figure 6-11 and the results in Table 1. For infrared images, whether the evaluation of indicators such as SSIM can be used for its quality evaluation needs to be considered.

If the author believes that the algorithm in this paper is effective, it is recommended to use it in the deblurring research of visible images.

Author Response

We would like to express our sincere gratitude to the reviewer for the detailed feedback and constructive suggestions, which have significantly improved the quality and clarity of our manuscript. We have carefully considered all the comments and made the necessary revisions to address the concerns raised. Below are our responses to the specific points highlighted.

1. Regarding the deblurring problem of infrared remote sensing images: Thank you for your insightful comments on the blurring problem of infrared remote sensing images. We fully agree that atmospheric effects are typically corrected using atmospheric correction methods. However, the blurring of infrared images is not solely caused by atmospheric effects; factors such as detector vibrations, defocus, and misfocus are also common causes of blurring. Atmospheric correction mainly addresses blurring before the sensor receives the image, while our proposed blind deblurring method targets the blurring caused by factors such as detector vibrations, defocus, and misfocus, which occur after image capture. Therefore, the blurring of infrared images is not limited to focus-related or low-resolution blurring; its sources are more complex. Nevertheless, we believe that the blurring of infrared remote sensing images still significantly affects image quality in practical applications, especially in remote sensing contexts. Therefore, studying blind deblurring for infrared remote sensing images remains meaningful. As the resolution of infrared images improves, many deblurring methods effective for visible light images can also be applied to infrared images, further enhancing the research value of deblurring methods. In fact, similar TV regularization-based deblurring methods have shown success in infrared remote sensing images. For example, in the paper "A Combined Stripe Noise Removal and Deblurring Recovering Method for Thermal Infrared Remote Sensing Images," a TV regularization method effectively handled deblurring of infrared remote sensing images. We agree with your suggestion to approach this research from a practical application perspective and will explore this issue further in the future, integrating our method with real-world remote sensing applications to enhance its practical value.
2. Regarding the visualization of experimental results and the applicability of evaluation metrics: We fully understand your concerns regarding the visualization of experimental results and the applicability of evaluation metrics. Indeed, traditional evaluation metrics such as SSIM may have certain limitations in assessing the quality of infrared images, as they differ from visible light images in their characteristics. To provide a more comprehensive assessment of our algorithm's performance, we have included additional evaluation metrics suitable for infrared images, such as NIQE, EME, and image entropy. These metrics offer a more nuanced reflection of local image information richness and blurriness and help to demonstrate the performance of different methods based on these indicators. We will continue to refine and explore more appropriate evaluation standards for infrared images in our future work.
3. Regarding the suggestion to apply this method to visible light images: We appreciate the reviewer's suggestion to apply our method to visible light images, and we strongly agree. We plan to evaluate the effectiveness of this method on visible light images in future work. We believe that this method holds significant potential for visible light image applications, and we will continue to explore its effectiveness in that domain.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript addresses an important problem in the field of infrared remote sensing by proposing a novel blind deblurring model. The integration of composite gradient feature (CGF) is innovative. However, there are several areas that could be further clarified or refined for better comprehension and impact.

1)       The paper is well-structured, with a logical flow of ideas from introduction to experiments. However, some sections, especially the theoretical derivations and algorithmic explanations, are dense and could benefit from simplification or additional explanations to enhance readability.

2)       The authors should emphasize more clearly how their approach differs from existing methods, particularly in terms of its real-world applicability and performance benefits.

3)       The main contribution part is somewhat misaligned, with too much emphasis placed on describing the experimental results. The true contributions of the work are not clearly highlighted and should be explicitly stated to underscore the novel aspects of the research.

4)       Some images in Figure 1 are not clearly visible, so consider improving their clarity and resolution for better comprehension. Figures 12 and 14 lack clarity, and the naming in 12(a), 12(b), and 14(a) is inconsistent with the algorithms. To enhance the comparison, consider adding a line for "ours without CGF" in Figure 12(b), which would provide a clearer illustration of the method’s advantages.

Author Response

We would like to express our sincere gratitude to the reviewer for the detailed feedback and constructive suggestions, which have significantly improved the quality and clarity of our manuscript. We have carefully considered all the comments and made the necessary revisions to address the concerns raised. Below are our responses to the specific points highlighted.

1. Regarding the simplification of theoretical derivations and algorithm explanations: Thank you for your valuable suggestion. We have simplified the theoretical derivations and algorithm explanations, reducing redundant explanations and adding necessary clarifications to enhance readability and understanding.
2. Regarding emphasizing the differences between our approach and existing methods, and its real-world applicability: Following your suggestion, we have more clearly highlighted the differences between our approach and existing methods in the abstract and main contributions section, particularly in terms of its real-world applicability and performance benefits. We have provided a detailed explanation of the uniqueness of our method and emphasized its potential value in practical problems.
3. Regarding the main contribution section: We have rewritten the main contributions section to more clearly highlight the innovation of this research and explicitly state the unique contributions of our work in both the algorithm and its applications. We believe this revision better reflects the core value of the paper.
4. Regarding image quality and naming improvements: Thank you for pointing out the clarity issues with some images. We have redrawn all unclear images to ensure their resolution and readability have been improved. Additionally, we have corrected the naming issues in Figures 12 and 14 to ensure consistency with the algorithms represented. As per your suggestion, we have also added a line for "ours without CGF" in Figure 12(b), now updated to Figure 13(b), to provide a clearer illustration of the method's advantages.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

1. It is suggested that the author provides a more detailed description of the image degradation model based on the Jilin-1 satellite mentioned on lines 583 and 584, so that readers can comprehensively understand the construction and types of degradations executed by this model.

2. The author mentioned that noise has a significant impact on deblurring algorithms. However, the blurred images used in the algorithm validation of this paper do not explicitly mention the introduction of noise. To fully assess the robustness of the proposed algorithm under noisy interference, it is recommended that the author introduces various types and levels of noise during the validation process and demonstrates the performance of the algorithm under these conditions.

3. It is recommended that, in addition to including methods based on the Total Variation (TV) model, the author also considers incorporating other deblurring techniques specifically designed for infrared images in the comparative analysis. Such a comparison would more comprehensively reflect the advantages of this paper's algorithm over existing technologies, especially when dealing with infrared images that have special characteristics.

4. To more intuitively demonstrate the advantages of this paper's algorithm in detail recovery, it is suggested that the author provide more magnified image details in the results presentation section. Currently, only a few images show magnified views, and the unmagnified images do not visually convey the advantages over other algorithms.

Author Response

We would like to express our sincere gratitude to the reviewer for the detailed feedback and constructive suggestions, which have significantly improved the quality and clarity of our manuscript. We have carefully considered all the comments and made the necessary revisions to address the concerns raised. Below are our responses to the specific points highlighted.

1. Regarding the detailed description of the image degradation model: Thank you for your suggestion. We have provided a more detailed description of the image degradation model in lines 583 and 584 of the paper. The experimental images are generated by convolving the original images with different Gaussian blur kernels. To address the image degradation in different complex scenarios and fully validate the robustness of our algorithm, we varied the kernel size and the variance in both directions for each image.
2. Regarding the introduction of noise robustness experiments: Following your suggestion, we have added noise robustness experiments in Section 6.7 of the paper. In these experiments, we introduced Gaussian noise and salt-and-pepper noise with different densities and amplitudes into the blurred images. We then computed the SSIM similarity between the blur kernels estimated by various methods and the true blur kernel in the degradation process, to demonstrate the robustness of each method under noisy interference.
3. Regarding the expansion of comparative methods: We have incorporated algorithms specifically designed for deblurring infrared remote sensing images into the comparative analysis, such as "A Combined Stripe Noise Removal and Deblurring Recovering Method for Thermal Infrared Remote Sensing Images" and "Fast Motion-Deblurring of IR Images." In addition, we have included other types of blind deblurring algorithms, such as "Image Deblurring Using Tri-Segment Intensity Prior." These added methods help to more comprehensively demonstrate the advantages of our approach, particularly in infrared image processing.
4. Regarding the visualization of detail recovery: Following your suggestion, we have magnified the comparison results for six sets of experiments to clearly show the restoration of image details and textures. This provides a more intuitive demonstration of the advantages of our algorithm in detail recovery.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I think the author has fully considered the questions I raised and made adequate replies and revisions. Therefore, I have reason to believe that the manuscript already has the conditions for acceptance by "remote sensing".