High-Precision Satellite Video Stabilization Method Based on ED-RANSAC Operator

Round 1

Reviewer 1 Report

The proposed manuscript presents a high-precision video stabilization method based on Euclidean distance RANdom SAmple Consensus (RANSAC) algorithm. RANSAC is a common algorithm used for robust estimation of parameters from a set of data points that contains outliers or noise. The image matching method used for the verificaton of accuracy is based on comparing feature distances on every 10th frame and checking accumulated error. The experiments were conducted using optical video data from the Jilin-01 satellite and airborne SAR video data. Some quite good results are reported. 

The paper is well structured and contains all the essential elements, but there are a few minor things that need to be corrected. What does the term "name" mean in the context of points? The term "name" appears only in Fig. 2 ("same name points"). It should be defined to understand the figure correctly. The terms "ED value" and "D-value" should also be defined in the paper, as well as the term "RANSAC operator".

The term "chi-square coordinates", is not a standard term or concept. Here it is used for homogeneous coordinates, or something similar. The chi-square test is a statistical test used to determine if there is a significant association between two categorical variables.

Author Response

Dear Reviewers,

 

Thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits. We have addressed all the comments point by point. In the revised manuscript, newly added or modified parts are indicated in red or revision mode.

 

Comments to Authors:

The proposed manuscript presents a high-precision video stabilization method based on Euclidean distance Random Sample Consensus (RANSAC) algorithm. RANSAC is a common algorithm used for robust estimation of parameters from a set of data points that contains outliers or noise. The image matching method used for the verification of accuracy is based on comparing feature distances on every 10th frame and checking accumulated error. The experiments were conducted using optical video data from the Jilin-01 satellite and airborne SAR video data. Some quite good results are reported.

 

The paper is well structured and contains all the essential elements, but there are a few minor things that need to be corrected. What does the term "name" mean in the context of points? The term "name" appears only in Fig. 2 ("same name points"). It should be defined to understand the figure correctly. The terms "ED value" and "D-value" should also be defined in the paper, as well as the term "RANSAC operator".

 

Response:

The layout of this manuscript has been reorganized and all figures and tables have been inserted into the paper.

1. The "same name points" in Figure 2 means the homonymous points. Due to negligence, we were not consistent with the description in the text. Thank you for your discovery and correction. We have been modified and inserted into the text. The results of the image modification are as follows:

Figure 2. ED-RANSAC algorithm flow chart.

2. In the description of Figure 4, we explained ED-Value: ED-Value is the threshold value used for secondary screening of the homologous points. Change as follows (additions are bolded and underlined here):

Figure 4. ED value verification chart. The relationship between the ED value and RMSE is shown on the left side of the image axis, and the relationship between the ED value and the Correct Matching Number (CMN) is established on the right side. ED-Value is the threshold value used for secondary screening of the homologous points.

 

3. D-Value is the difference between the inter-frame RMSE and the average of the total RMSE. It is used to evaluate the volatility of the algorithm. The changes are as follows (additions are bolded and underlined here to show, and Figure 10 changes are the same as Figure 6, which is used as an example below):

Figure 6. ED-RANSAC algorithm stability analysis (left) and comparison with RANSAC, LO-RANSAC and 3sigma stability image results (right). From top to bottom: Zhifu Bay in Yantai, Jiayuguan in Gansu, Leibo County in Sichuan. D-Value is the difference between the inter-frame RMSE and the average of the total RMSE.

 

4. The "RANSAC operator" is the name of the operator proposed in this paper, which is the "ED-RANSAC operator" in the title of this paper. It is explained in line 5 of the abstract section:

This paper proposed a high-precision VSM based on the ED-RANSAC, an error elimination operator constrained by Euclidean distance.

 

The term "chi-square coordinates", is not a standard term or concept. Here it is used for homogeneous coordinates, or something similar. The chi-square test is a statistical test used to determine if there is a significant association between two categorical variables.

 

Response:

The original meaning of "chi-square coordinates" mentioned in this article is "homogeneous coordinates", and writing "chi-square coordinates" is an expression error, which brings inconvenience to your review. We have corrected it to "homogeneous coordinate". Homogeneous coordinates use N+1 dimensions to represent N-dimensional coordinates. Using homogeneous coordinates can better represent the position of points in 2D space in perspective space. It can be used to clearly distinguish vectors and points, and it is also easier to use for affine (linear) geometric transformations.

 

Before modification:

Eq. (5) represents the two-dimensional plane coordinates (x_s, y_s) obtained after the transformation from the original coordinates consisting of the chi-square coordinates (x, y,1).

 

After modification:

Eq. (5) represents the two-dimensional plane coordinates (x_s, y_s) obtained after the transformation from the original coordinates consisting of the homogeneous coordinate (x, y,1).

 

 

 

We would like to take this opportunity to thank you for all your time involved and this great opportunity for us to improve the manuscript. We hope you will find this revised version satisfactory.

 

Sincerely,

The Authors

Author Response File: Author Response.docx

Reviewer 2 Report

This manuscript presents a high-precision satellite video stabilization method for optical and SAR video data. The proposed method, called ED-RANSAC, utilizes Euclidean distance to enable precise extraction of homologous features. The experiments show that the approach can achieve great stabilization performance for both optical and SAR data. Nevertheless, there are several issues that need to be addressed:

 

1.     The ‘motion’ part in Figure 1 is confusing, as it appears to depict only transformation methods being used in the process without incorporating multi-frame motion estimation, such as optical flow. Additionally, on page 9, line 9, the term 'inter-frame transformation model' is used; I recommend using the term 'transformation model' instead.

2.     Regarding point selection, it is worth noting that the gap between homologous point pairs in the three scenarios is relatively large, with values of 19599, 18197, and 12169, respectively. This large number of points can pose a significant operational burden for RANSAC. However, by utilizing Euclidean distance, it may be possible to reduce the computational cost. It would be interesting to investigate how many points Euclidean distance could screen out and how this could impact the computational efficiency of the ED-RANSAC method.

3.     In Figure 5, the line chart on the right to be somewhat unintuitive. To improve the clarity of the chart, I recommend changing the line colors and highlighting the best results. In Figure 6, I suggest using thicker lines to improve the visibility and clarity of the plotted data.

4.     To provide a more comprehensive comparison, I suggest including additional contrastive methods in addition to RANSAC. Regarding the accuracy evaluation process, it would be important to carefully select the same homologous points across different methods to ensure a fair and accurate comparison. Further details on the point selection process could be provided to enhance the transparency and rigor of the evaluation.

5.     In Figure 8 and 12, it may be more convincing to use frames that are far apart rather than adjacent frames. Additionally, to better illustrate the performance, I recommend including the original, unstabilized frames as a reference in the figure.

6.     The method section of the article is short, and it is recommended to supplement and adjust content to demonstrate the innovation.

7.     There are many formatting issues in the paper, such as the position of the formula's sequence number, the capitalization of the first letter in Keywords, and abbreviations in Introduction (RPC repeatedly provides abbreviation annotations; DEM does not provide full names), etc.

8.   Some expressions are redundant and inappropriate, for example, on Page 12 line 1, the phrase 'To visually evaluate' could be changed to 'To visualize' to eliminate redundancy and improve clarity. Similarly, on Page 11 line 15, using the phrase 'every 10 frames' instead of 'every 10th frame' would be more precise and appropriate.

Minor editing of English language required

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have well addressed my comments. I recommend this paper for publication.