Analysis of Stochastic Distances and Wishart Mixture Models Applied on PolSAR Images
Round 1
Reviewer 1 Report
This paper analyzes the Stochastic Distances and Wishart Mixture Models for PolSAR images using the clustering task. Although the idea is feasible, there are many issues should be noticed before publishing.
The main contributions and motivation are not clear enough. From the current version, readers can only learn that authors do some clustering experiments with the traditional algorithms. The definition of assessment criteria should be provided. In addition, more assessment criteria should be added to study your main idea. The comparisons should be enriched. The latest methods should be selected. Written English should be improved. There are many grammar errors, typos, and improper statements in the current manuscript. Reference should be enriched.Author Response
Dear Reviewer ,
I would like to thank you for your letter and the opportunity to revise our article on 'Analysis of Stochastic Distances and Wishart Mixture Models applied on PolSAR images.' The suggestions offered by you have been immensely helpful.
We added the lines 243 to 247 and lines 426 to 432 to explain number of looks importance for Wishart distribution and the stochastic distance, and for the success of our classification.
The explanation about why SC-C method has such a poor performance and why it seems to have so many outliers is described in lines 309 to 315.
Regarding English writing, we have reviewed the article and corrected the typos and grammar mistakes. We apologise for neglecting the writing when we originally submitted the manuscript.
We hope the revised manuscript will better suit the Remote Sensing Journal, but are happy to consider further revisions, and we thank you for your continued interest in our research.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper presents an interesting comparison between different classification strategies applied in Polarimetric SAR images. The paper presents some interesting findings and I only have minor comments that can help the authors improve the quality of the paper for it to suitable for publication.
Minor comments
What is the impact of the number of looks in the performance of the different algorithms?. Although from a clutter level is obvious why the authors simulated a multilook image, it is interesting that lower resolution images were used as a testbed. Do the authors have any potential hypothesis why the SC-C method seems to have so many outliers?.Author Response
Dear Reviewer ,
I would like to thank you for your letter and the opportunity to revise our article on 'Analysis of Stochastic Distances and Wishart Mixture Models applied on PolSAR images.' The suggestions offered by you have been immensely helpful.
I have change the abstract to make it clear, and tried to clarify the our work main contributions and motivation. Regarding the algorithms evaluation criteria, we added the overall accuracy equation (page 10, Equation 23). Also, we added confusion matrices graphs for each classification algorithm. We added discussions about the results (Lines 287 to 314) and we describe the behavior of each stochastic distances.
Regarding English writing, we have reviewed the article and corrected the typos and grammar mistakes. We apologise for neglecting the writing when we originally submitted the manuscript.
We hope the revised manuscript will better suit the Remote Sensing Journal, but are happy to consider further revisions, and we thank you for your continued interest in our research.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper analyzes various methods of unsupervised classification; special attention is paid to comparing different distances in terms of classification accuracy.
In general, the content is presented sequentially and logically, the conclusions are supported by the results of the classification of surfaces, both simulated and real. The language of the paper is good, but there are a number of inaccuracies and small errors, therefore, I suggest making a final proofreading (for example, there are errors associated with matching up singular and plural forms).
The paper does not contain much novelty, but may be useful for researchers choosing the unsupervised classification method.
The paper presents in sufficient detail the theoretical foundations underlying the methods under consideration, as well as the results of comparing methods for accuracy. However, an important practical issue of computational cost of implementing these methods is not considered in the paper. If the costs are approximately the same, then this should be mentioned.
I have some comments on the paper:
Fig. 3 - it should be clarified which of the figures relates to strong and which refers to weak signal.
Line 200. Which criteria used for comparison of surface roughness with the radar wavelength?
Line 242 (Fig. 6). When the authors say that “these particular results show better performance ... over the traditional method”, then this does not quite correspond to Fig. 6, where some results are much weaker than in the KM-ED method.
Line 261-273. I agree with the authors that the accuracy of the classification depends on the choice of initial centroids. However, the authors do not say anything about how to avoid adverse cases in choosing initial centroids.
The paper lacks analysis of why classification when using some methods (distances) shows better results than when using others. Only the results are given. Such an analysis would allow us to understand how it is possible to extend the results to other areas of classification, which would make the paper universal and results more valuable.
Author Response
Dear Reviewer,
I would like to thank you for your letter and the opportunity to revise our article on 'Analysis of Stochastic Distances and Wishart Mixture Models applied on PolSAR images.' The suggestions offered by you have been immensely helpful.
We splitted the Fig. 3 into Fig. 3 a), Fig. 3 b) and Fig. 3 c. Hopefully it will be more clear about which figure refers to strong and which refers to weak signal.
About the criteria used for comparison of surface roughness with the radar wavelength, it is the Rayleigh Criterion, which is described in lines 226-236.
In Fig. 6, you are right, the SC-R and SC-C are much weaker than in the KM-ED method. I corrected this paragraph. You can check on lines 288-292.
Regarding the dependency of classification accuracy on the choice of initial centroids, in lines 422-425 you can check some initialization algorithms examples.
In results we added an explanation about the stochastic distance difference of performance. For instance, the why SC-C method has such a poor performance and why it seems to have so many outliers is described in lines 309 to 315.
Regarding English writing, we have reviewed the article and corrected the typos and grammar mistakes. We apologise for neglecting the writing when we originally submitted the manuscript.
We hope the revised manuscript will better suit the Remote Sensing Journal, but are happy to consider further revisions, and we thank you for your continued interest in our research.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have done a great job and the new version of the paper looks much better than the first. All my comments have been taken into account and the corresponding changes have been made. Some drawings have been improved as well as the language of the paper.
