A Multi-Feature Framework for Quantifying Information Content of Optical Remote Sensing Imagery
Round 1
Reviewer 1 Report
Figure 1 should be converted to a better resolution. Details are not clearly visible. Also it would be better to adjust the figure for the use-case approached in the experiments: conversion of the color image to grayscale.
It is not clear what is the meaning of variable nnei(i,j) from eq. 4. Is it the pixel value of the j neighbor ? In that case, Hnei is a kind of average of the pixel values for the considered image channel, so nothing in common with the contrast of the channel.
Is not clear how the values from the examples given figure 5 are computed. Hence the meaning of eq. 5 is hard to understand.
Is not clear how eq. 7 is computed for a multi-spectral / multichannel images. The summation is done at each information feature H or the final H-values of each channels are summed?
For the plots from figure 10 is not clear what is the meaning of the horizontal axis (down-sampling ratio ?).
Regarding the experimental results presented in section 4, results for experiments 2 and 3 are straightforward to characterize the invariance relative to geometrical transformations of the proposed information/entropy measures in terms of scale/resolution (experiment 2) and translation (experiment 3).
Interpretation of the results in experiment 1 are image content dependent and only are showing the information/entropy differences between the proposed method and the compared methods from the literature for a particular/reduced set of images.
Also authors should state what is the applicability domain of the proposed method, and for which classes of images should be intended to be used. It is not clear the relevance of the proposed method end experiments relative to the topic of the Journal/issue (Remote Sensing and Geo-Spatial Science).
Regarding experiment 1, I may suggest that a machine learning approach in which the results are compared/classified over a set of relevant user labeled image classes could be more relevant. In that case, instead of summing the H components in a scalar value as in equation 7, a feature vector of dimension 4 could be used instead. Obviously for the comparison with the existing methods from the literature, the information measure provided by them will be classified in a 1 dimensional feature space over the same labelled data set..
Author Response
Dear Editor and Reviewers,
Thank you very much for comments and constructive suggestions from reviewers with regard to our manuscript " A Multi-feature Framework for Quantifying Information Content of Optical Remote Sensing Imagery " (remotesensing-1833822). No conflict of interest exists in the submission of this manuscript, and manuscript is approved by all authors for publication. The paper is our original unpublished work and not under consideration for publication elsewhere.
Based on the suggestions of the reviewers, we have revised Figure 1, Figure 3 and Figure 13, added explanations and descriptions of figures and formulas in the manuscript to ensure that the meaning of the details is clear enough, rephrased the literature review, corrected minor grammar and typos in the manuscript. etc. Besides, we have corrected the format of all references and other formats as required by instructions for Remote Sensing authors.
We are uploading (a) our point-by-point response to the comments (below) (response to reviewers), (b) an updated manuscript with yellow highlighting indicating changes.
We would like to express our great appreciation to you and reviewers for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Luo Silong et al.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Remote Sensing.
Review Report
Manuscript ID: remotesensing-1833822. "A Multi-feature Framework for Quantifying Information Content of Optical Remote Sensing Imagery"
This manuscript is about a very interesting study that addresses a topical topic, is well structured, very well presented, and proposes a method to quantify multiple characteristics of the information content of images based on grayscale, contrast, topology based on the neighborhood and spatial distribution characteristics; the entropy, the uncertainty in terms of pixel and spatial structure, respectively.
I find very few areas of opportunity to improve, which I put into consideration with the authors
COMMENTS
1 The work states that the Shannon entropy model will no longer be available, and other information quantification models need to be constructed instead. Before that, it is mentioned that the techniques used to evaluate the shapes and spatial relationships in an image inevitably introduce uncertainty, breaking the closure of the original system. Furthermore, something is also mentioned in topic 2.1 and lines 534-538.
I understand that Shannon's model has been around for a long time. However, I have doubts if these reasons explained by the authors are sufficient to affirm that this model will no longer be available.
It is just a doubt; maybe the fact that Shannon entropy relies entirely on the statistics of pixel grayscale and fails to capture the spatial structure of imagery is enough.
2 Review questions of fo Review questions of the form (not substance) in writing, such as the case of repeated words, for example
Shannon's information theory quantifies the content of information content. (word content repeated in line 175)
3 Throughout the document, particularly topic 3, and according to the explanations, I wondered how to implement it in the case of multi-band images. There is a brief explanation of it (lines 334-338), where the authors indicate that it should be noted that remote sensing imagery has multiple bands; without considering the correlation of bands, the information of multi-band images can take the sum value of the information content of each band. The RGB images were converted to grayscale images for information measurement in our experiments.
So this explanation means that for a multi-band image, I must add the individual values of each band ​​to obtain a global image and apply the proposed method to this global image. It is, right?
My suggestion here is to emphasize this information. If it is possible to indicate it before, it can be in the initial paragraphs of topic three and not at the end because throughout topic three is where in my case, the doubt arises as to how to treat a multi-band image.
4. Line 366 mentions Wasserstein-based Boltzmann entropy with von Neumann neighborhood (WEH4) [20]… the citation refers to (Neumann, 1996). Just confirm whether it is Jan Neuman or von Neumann, as the document says.
Author Response
Dear Editor and Reviewers,
Thank you very much for comments and constructive suggestions from reviewers with regard to our manuscript " A Multi-feature Framework for Quantifying Information Content of Optical Remote Sensing Imagery " (remotesensing-1833822). No conflict of interest exists in the submission of this manuscript, and manuscript is approved by all authors for publication. The paper is our original unpublished work and not under consideration for publication elsewhere.
Based on the suggestions of the reviewers, we have revised Figure 1, Figure 3 and Figure 13, added explanations and descriptions of figures and formulas in the manuscript to ensure that the meaning of the details is clear enough, rephrased the literature review, corrected minor grammar and typos in the manuscript. etc. Besides, we have corrected the format of all references and other formats as required by instructions for Remote Sensing authors.
We are uploading (a) our point-by-point response to the comments (below) (response to reviewers), (b) an updated manuscript with yellow highlighting indicating changes.
We would like to express our great appreciation to you and reviewers for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Luo Silong et al.
Author Response File: Author Response.docx
Reviewer 3 Report
This paper developed a multi-channel framework to extract information from remote sensing images. It considers the grayscale, neighborhood information, and spatial features simultaneously. Overall, this paper is technically sound. However, there are sill some issues that need revision at current stage:
First, this framework indeed extract features with more information. However, this main problem is that what is the utility of the extracted features? This is related to the purpose of this study and is vague in this paper.
Second, the literature review is very weak. A lot of recent studies are not mentioned.
More, figure 3 is at low quality. Please re-draw the figure with higher resolution.
In addition, in Table 2, it seems that all information content except the spatial information is identical among the original and simulated image. Does it mean that the proposed framework only enhanced the information extraction of spatial features but nothing else? Please discuss.
Last, please use a new figure containing more simulated image. Then the reviewer and readers can visually compared the original image and the simulated image which is important for demonstrating the utility of this study.
After revision, this paper has the potential for being considered for publication.
Author Response
Dear Editor and Reviewers,
Thank you very much for comments and constructive suggestions from reviewers with regard to our manuscript " A Multi-feature Framework for Quantifying Information Content of Optical Remote Sensing Imagery " (remotesensing-1833822). No conflict of interest exists in the submission of this manuscript, and manuscript is approved by all authors for publication. The paper is our original unpublished work and not under consideration for publication elsewhere.
Based on the suggestions of the reviewers, we have revised Figure 1, Figure 3 and Figure 13, added explanations and descriptions of figures and formulas in the manuscript to ensure that the meaning of the details is clear enough, rephrased the literature review, corrected minor grammar and typos in the manuscript. etc. Besides, we have corrected the format of all references and other formats as required by instructions for Remote Sensing authors.
We are uploading (a) our point-by-point response to the comments (below) (response to reviewers), (b) an updated manuscript with yellow highlighting indicating changes.
We would like to express our great appreciation to you and reviewers for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Luo Silong et al.
Author Response File: Author Response.docx
Reviewer 4 Report
Comments to Authors:
The work proposes a multi-feature approach for measuring of image information content in remote sensing images. The authors defined a comprehensive information content of the image as a weighted sum of the following Shannon's entropy metrics: grayscale, contrast, neighborhood-based topology and spatial distribution features, instead of simple grayscale or spatial structure metrics. They showed that calculated image information is consistent with the human visual perception of information. Undoubtedly, further research is required on the applicability of the proposed approach for estimating information in optical remote sensing images.
However, I have a number of comments both on the content and on the style of the text.
Some specific comments.
1. Line 175: two times “content”.
2. Lines 212 – 218: unconvincing reasoning.
3. Eq. (7): It is not clear to me what is the practical meaning of adding information of different nature in order to obtain general information. I believe, for example, that a person does not perceive grayscale entropy.
4. Line 550: defining -> define.
5. Line 570 – 571: the phrase is strange, it seems there is no predicate in it.
Author Response
Dear Editor and Reviewers,
Thank you very much for comments and constructive suggestions from reviewers with regard to our manuscript " A Multi-feature Framework for Quantifying Information Content of Optical Remote Sensing Imagery " (remotesensing-1833822). No conflict of interest exists in the submission of this manuscript, and manuscript is approved by all authors for publication. The paper is our original unpublished work and not under consideration for publication elsewhere.
Based on the suggestions of the reviewers, we have revised Figure 1, Figure 3 and Figure 13, added explanations and descriptions of figures and formulas in the manuscript to ensure that the meaning of the details is clear enough, rephrased the literature review, corrected minor grammar and typos in the manuscript. etc. Besides, we have corrected the format of all references and other formats as required by instructions for Remote Sensing authors.
We are uploading (a) our point-by-point response to the comments (below) (response to reviewers), (b) an updated manuscript with yellow highlighting indicating changes.
We would like to express our great appreciation to you and reviewers for comments on our paper. Looking forward to hearing from you.
Thank you and best regards.
Luo Silong et al.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors have improved the quality of the paper and have addressed all the issues mentioned in the first review report.
Reviewer 3 Report
This paper is fine for now
