Analysis of Despeckling Filters Using Ratio Images and Divergence Measurement

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper proposed a SAR noise speckle suppression analysis that combines quantitative evaluation indicators and probability distribution of ratio images for divergence. The experimental results indicate that the analysis has certain value. However, there are still some problems to be considered by the authors. My main comments are as follows:

(1)   In the experiment of optical images corrupted by speckle in Section 4.2, in the region of interest shown in Experiment 4 of Figure 3, the smoothness of the results of FANS filtering is significantly better than that of ELee and Lee. Why is the FANS index worse than these two filtering methods in the ENL index?

(2)   The selection of indicators in this article focuses on the unilateral evaluation of speckle suppression and edge preservation. The author can refer to some comprehensive evaluation indicators [1-2].

(3)   From the ratio image shown in Figure 7, all other filtering methods have obvious residual information, except for SCUNet. However, in the edge preservation index, SCUNet did not show good performance. Please provide a reasonable explanation.

(4)   In Table 12, a JSD index of 0 for the Lee method indicates perfect filtering, but in Figure 6, there are still granular spots present. Please provide a reasonable explanation.

(5)   In Table 6, the JSD index of Lee's method performed the best in 1, 3, and 4 images, but Lee did not show significant performance in speckle suppression or edge preservation. Please provide a reasonable explanation.

[1] L. Gomez, M. E. Buemi, J. C. Jacobo-Berlles and M. E. Mejail, "A New Image Quality Index for Objectively Evaluating Despeckling Filtering in SAR Images," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 3, pp. 1297-1307, March 2016, doi: 10.1109/JSTARS.2015.2465167.

[2] Gomez, L.; Ospina, R.; Frery, A.C. Unassisted Quantitative Evaluation of Despeckling Filters. Remote Sens. 2017, 9, 389. https://doi.org/10.3390/rs9040389

Comments for author File: Comments.pdf

Author Response

Comments 1: In the experiment of optical images corrupted by speckle in Section 4.2, in the region of interest shown in Experiment 4 of Figure 3, the smoothness of the results of FANS filtering is significantly better than that of ELee and Lee. Why is the FANS index worse than these two filtering methods in the ENL index?

Response 1: Thanks for the comment, we have run the experiments again in order to confirm the results. The quantitative result is the same as included in the original manuscript. The explanation of this is due to the 20x20 region of interest used to measure the ENL is homogeneous and it is different from the heterogeneous region used to zoom the image in the paper. This was intentional, since the former choice was for measurement purposes, while the latter is for illustrative purposes and visual inspection in the article.

Comments 2: The selection of indicators in this article focuses on the unilateral evaluation of speckle suppression and edge preservation. The author can refer to some comprehensive evaluation indicators [1-2]. [1] L. Gomez, M. E. Buemi, J. C. Jacobo-Berlles and M. E. Mejail, "A New Image Quality Index for Objectively Evaluating Despeckling Filtering in SAR Images," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 3, pp. 1297-1307, March 2016, doi: 10.1109/JSTARS.2015.2465167. [2] Gomez, L.; Ospina, R.; Frery, A.C. Unassisted Quantitative Evaluation of Despeckling Filters. Remote Sens. 2017, 9, 389. https://doi.org/10.3390/rs9040389

Response 2: Thanks for the recommendation. Both references were included and described in the introduction. For edge preservation measurement, the beta correlator was included also in the introduction and in chapter 2, as part of the measurements considered to assess the despeckling quality. Tables and results were included in the rest of the chapters. PFOM is another measurement of edge preservation that was included from the first version of the manuscript.

Comments 3: From the ratio image shown in Figure 7, all other filtering methods have obvious residual information, except for SCUNet. However, in the edge preservation index, SCUNet did not show good performance. Please provide a reasonable explanation. Response 3: Thanks for the recommendation. As shown in Table 8 (ENL), Table 9 (MSE), Table 10 (SSIM), Table 11 (PSNR), newly added Table 12 (Beta) and Table 13 (PFOM), the SCUNet model poorly suppressed the speckle present in the SAR images used in the paper. Table 14 (JSD), which was built by using the ratio images, confirms these findings when the JS distances are higher when using this model, which yields to the conclusion that the noise shown in Figure 7 for this model is very different from the original speckle in the SAR image. This ratio image contains high residual speckle and the mean was not preserved. The ratio analysis proposed in the manuscript is useful as a second order measurement. In the scenario of a bad filtering process, the indexes and the ratio analysis are not suitable. We recommend the use of this methodology only after a first order evaluation of the despeckling process. We implemented some modifications with this analysis in the newly created Section 5.3. Comments 4: In Table 12, a JSD index of 0 for the Lee method indicates perfect filtering, but in Figure 6, there are still granular spots present. Please provide a reasonable explanation. Response 4: The JSD proposed to analyze ratio images in this paper is not intended to be the unique measurement of a despeckling process, so a value of 0 in the JSD does not mean a perfect filtering. All the indexes used in this manuscript, proposed for several great authors, have limitations and must be used together to assess the despeckling quality. Some of them measure the edge preservation, the remaining geometry, and for instance SSIM is a global measurement while the MSE is a pixel by pixel comparison. In conclusion, there is no perfect filter and so there is no perfect index to measure its performance. We, as authors of this manuscript, propose the JSD as one of the measurements that other authors can apply on filtered images, and in this case, ratio images, to assess its quality and find their own conclusions. Thanks for the recommendation. The manuscript was updated to include this analysis. Comments 5: In Table 6, the JSD index of Lee's method performed the best in 1, 3, and 4 images, but Lee did not show significant performance in speckle suppression or edge preservation. Please provide a reasonable explanation. Response 5:  Explanation of Lee Method's JSD Performance Table 6 is now changed to Table 7. This explanation is added in section 5.1. In Table 7, it is observed that the Jensen-Shannon Divergence (JSD) index of the Lee method performed the best in images 1, 3, and 4. However, the Lee method did not exhibit significant performance in terms of speckle suppression or edge preservation when compared to other filters. This discrepancy can be explained by the inherent characteristics of the JSD index and the nature of the Lee filter. The JSD index focuses on the statistical preservation of noise characteristics rather than visual quality. The Lee filter may excel in maintaining the statistical properties of the speckle noise, resulting in low JSD values. However, this does not necessarily translate to effective speckle suppression or edge preservation, which are critical for visual quality and structural integrity. The Lee filter's performance is particularly strong in homogeneous regions where it can accurately preserve the local statistics. In such areas, the ratio image's speckle remains close to the theoretical distribution, resulting in a lower JSD. This might explain its superior JSD performance in images 1, 3, and 4 if these images contain significant homogeneous regions. The Lee filter tends to blur edges and fine details because it averages pixel values within the local window. This limitation affects its performance in edge preservation and visual quality metrics like beta and PFOM, despite its ability to preserve the statistical properties of the speckle.  Thanks for the recommendation. The manuscript was updated to include this analysis.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript is useful for Synthetic Aperture Radar researchers and in order to to quantitatively evaluate the despeckling process in SAR images, various metrics were utilized. The authors proposed an evaluation of traditional and AI-based despeckling filters applied to both synthetically corrupted optical images and actual SAR images, using a novel method involving ratio images to measure residual speckle, i.e. divergence measurement approach to quantitatively and visually assess the similarity between initial and remaining speckle. Furthermore, Gamma distribution approximation is used in order to understand the behavior of residual speckle, offering new insights and tools for future research in SAR image despeckling.

Despeckling of images is used by six filters: Lee, ELee, and FANS are not AI- based, while the other three: MONET, AE, and SCUNet are DL-based. The resulting filtered images are evaluated by using well-known metrics: ENL, MSE, SSIM, PSNR, and PFOM. Finally, our proposal, is to assess the Gamma approximation of the resulting ratio of filtered image by comparing it with respect to the observed image (SAR or synthetically corrupted optical). 

The obtained results showed that across all evaluated metrics, FANS emerged as the most effective filtering method for reducing noise and preserving image fidelity and structural integrity in synthetically speckled optical images. SCUNet, while less effective across these metrics, highlights the variability in performance of different filtering methods. The AE filter, although not always the best in noise reduction, demonstrated strong performance in preserving important image features (patterns).

The study is well organized and deals with the narrow field of analysis of different despeckling filter, and and I can't find an objection.

Line 1: first time SAR is mentioned, maybe to define as Synthetic Aperture Radar?

 

Author Response

Comments 1: The manuscript is useful for Synthetic Aperture Radar researchers and in order to to quantitatively evaluate the despeckling process in SAR images, various metrics were utilized. The authors proposed an evaluation of traditional and AI-based despeckling filters applied to both synthetically corrupted optical images and actual SAR images, using a novel method involving ratio images to measure residual speckle, i.e. divergence measurement approach to quantitatively and visually assess the similarity between initial and remaining speckle. Furthermore, Gamma distribution approximation is used in order to understand the behavior of residual speckle, offering new insights and tools for future research in SAR image despeckling.

Despeckling of images is used by six filters: Lee, ELee, and FANS are not AI- based, while the other three: MONET, AE, and SCUNet are DL-based. The resulting filtered images are evaluated by using well-known metrics: ENL, MSE, SSIM, PSNR, and PFOM. Finally, our proposal, is to assess the Gamma approximation of the resulting ratio of filtered image by comparing it with respect to the observed image (SAR or synthetically corrupted optical).

The obtained results showed that across all evaluated metrics, FANS emerged as the most effective filtering method for reducing noise and preserving image fidelity and structural integrity in synthetically speckled optical images. SCUNet, while less effective across these metrics, highlights the variability in performance of different filtering methods. The AE filter, although not always the best in noise reduction, demonstrated strong performance in preserving important image features (patterns).

The study is well organized and deals with the narrow field of analysis of different despeckling filter, and and I can't find an objection.

Line 1: first time SAR is mentioned, maybe to define as Synthetic Aperture Radar?

 

Response 1: Thanks for the recommendation. The first time the term SAR is mentioned is in the abstract. The reviewer’s comment was considered in this place. The term is also defined in the keywords and at the end of the paper in the abbreviation section.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Your article provides a more robust framework for evaluating the effectiveness of despeckling techniques, particularly highlighting the flexibility and limitations of AI-based models when applied to real-world SAR imagery. Additionally, the study underscores the importance of gamma distribution approximation in understanding the behaviour of residual speckles, offering new insights and tools for future research in SAR image despeckling.

It is a well-structured article that may be of particular interest to those who are working on SAR images. In any case, I suggest that you make a series of modifications with a view to improving it. Specifically, I refer to:

-        Evaluate the methodology used in the study, commenting on its appropriateness and potential limitations.

-        Provide constructive feedback on the analysis conducted, pointing out any areas for improvement or further research.

-        Discuss the implications of the findings and how they contribute to the existing literature on despeckling filters.

- Based on the above, it is recommended that the authors review the entire document, taking into consideration the mentioned suggestions as well as any others that other reviewers may propose.

 

All the best.

Author Response

Response to Reviewer 3 Comments
1. Summary
Thank you very much for taking the time to review this manuscript, your comments help us improve it. We have considered and included all the recommendations in the new version of the manuscript. Please find the detailed responses below and the corresponding corrections highlighted in the re-submitted files.
2. Questions for General Evaluation	Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references?	Yes	The introduction section was reviewed and updated by including the other reviewer’s recommendations.
Are all the cited references relevant to the research?	Yes	The references were reviewed and updated by including the other reviewer’s recommendations.
Is the research design appropriate?	Yes	The design was reviewed and updated by including the other reviewer’s recommendations. Sections were added.
Are the methods adequately described?	Yes	The methods were reviewed and updated by including the other reviewer’s recommendations.
Are the results clearly presented?	Can be improved	The results section was updated and improved according to the reviewer’s suggestions.
Are the conclusions supported by the results?	Can be improved	The conclusion section was updated and improved according to the reviewer’s suggestions.
3. Point-by-point response to Comments and Suggestions for Authors
Comments 1: Your article provides a more robust framework for evaluating the effectiveness of despeckling techniques, particularly highlighting the flexibility and limitations of AI-based models when applied to real-world SAR imagery. Additionally, the study underscores the importance of gamma distribution approximation in understanding the behaviour of residual speckles, offering new insights and tools for future research in SAR image despeckling. It is a well-structured article that may be of particular interest to those who are working on SAR images. In any case, I suggest that you make a series of modifications with a view to improving it. Specifically, I refer to: Evaluate the methodology used in the study, commenting on its appropriateness and potential limitations.
Response 1: Thanks for the recommendation. The manuscript was updated to include this analysis. Section 5.4 was added. In section 5.4, the methodology used is evaluated, commenting on its appropriateness and potential limitations, the relevance and limitations of AI-based models are analyzed. Additionally, conclusions section has been expanded: New conclusion added to the manuscript: The methodology used in this study is robust and appropriate for evaluating the effectiveness of despeckling filters. While there are potential limitations, such as the synthetic nature of some tests and the assumptions in multitemporal fusion, the approach provides a comprehensive framework for assessing both traditional and AI-based despeckling techniques. Future research should aim to address these limitations by incorporating more diverse datasets and exploring additional evaluation metrics.

 

Comments 2: Provide constructive feedback on the analysis conducted, pointing out any areas for improvement or further research.

Response 2: We have added section 5.5 Feedback on Analysis. In this section, we provide constructive feedback on the analysis conducted, highlighting areas for improvement. In section 6.1 Future work, suggesting directions for future research. This ensures that the study not only evaluates the current results but also promotes ongoing development in the field of speckle suppression in SAR images.

 Thanks for the recommendation. The manuscript was updated to include this analysis.

Comments 3: Discuss the implications of the findings and how they contribute to the existing literature on despeckling filters.

Response 3: We have added section 5.6 Implications and Contributions to the Literature. In this section, we discuss the implications of the study's findings and how they contribute to the existing literature on despeckling filters. This provides a broader context and highlights the relevance and impact of the research in the field.

Thanks for the recommendation. The manuscript was updated to include this analysis.

Comments 4: Based on the above, it is recommended that the authors review the entire document, taking into consideration the mentioned suggestions as well as any others that other reviewers may propose.

Response 4: Thanks for the recommendation. The introduction was updated by including the Beta measurement according to other reviewer’s recommendations. All the analysis, conclusions and future work sections had to be updated, also from the previous comments of this 3rd reviewer.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All the issues I have raised have been improved.