Efficient and Robust Feature Matching for High-Resolution Satellite Stereos

Round 1

Reviewer 1 Report

The paper presents an image matching block selection algorithm for efficient and robust feature matching of satellite stereos to greatly reduce the matching time cost, while maintaining robust matching results, which is of practical interest for the satellite photogrammetry community. Both the algorithmic improvement and the experimental results support the contribution of this paper. Detailed comments are as follows:

 1. The English writing should be improved and grammatical errors must be corrected. For example:

- In page 2, line 76, assumptions -> assumption

- In page 2, line 92, “… while keeping matching accuracy of…” The authors should clarify which accuracy level will be kept.

- In page 3, line 107, potential -> candidate

- In page 5, line 208, … means finding the minimum … -> … means the minimum …

- In page 5, line 224, the greedy solution is always selecting … -> the greedy solution always selects …

- In page 6, line 255, the blocks sizes -> the block sizes

2. The authors have extended their experiments and evaluations to large satellite dataset, including WorldView and GaoFen-7. It is necessary to reference the following when using the CORE3D WorldView data (Omaha area), please see the link here, https://spacenet.ai/core3d/:

- M. Brown, H. Goldberg, K. Foster, A. Leichtman, S. Wang, S. Hagstrom, M. Bosch, and S. Almes, “Large–Scale Public Lidar and Satellite Image Data Set for Urban Semantic Labeling,” in Proc. SPIE Laser Radar Technology and Applications XXII, 2018.

- Commercial satellite imagery in the CORE3D public dataset was provided courtesy of DigitalGlobe.

- Dataset was created for the IARPA CORE3D program: https://www.iarpa.gov/index.php/research-programs/core3d.

- SpaceNet on Amazon Web Services (AWS). “Datasets.” The SpaceNet Catalog. Last modified October 15, 2018. Accessed on [Insert Date]. https://spacenetchallenge.github.io/datasets/datasetHomePage.html.

3. In Line 104, when creating the pyramid, why is the zoomed-out ratio set to 4 for the computation efficiency purpose, not the other values? While in Line 150, the scaling factor s is defined as 3? The authors should explain why the above parameters are set in such values, respectively.

4. In Line 107, what is the size of the divided blocks? Are there overlap regions among the divided fixed blocks? For it is commonly that texture-aware regions might be divided into two or more blocks on the top pyramid images, will that cause failure of your approach? 

5. In the second-step optimization, how to obtain the common height plane?

6. In the first-step optimization, is it possible to select the blocks beyond the overlaps of the stereos?

7. In the post-processing of the proposed method, the authors said “the block sizes in both I1 and I2 are similar”. However, When the GSD of I1 and I2 are significantly different, the corresponding block sizes should be adjusted to their GSD, which results in different blocks sizes. How did the authors address this issue?

8. The elements (e.g. set, matrix, vector, scalar) in all equations have similar formats with italic and normal fonts, which may bring difficulties for readers in understanding the equations. It is recommended to apply different fonts to different elements, which can refer to the author guidance of RS.

9. In page 6, the authors should explain the meanings of the red rectangles and the green rectangle in Figure 2.

10. The format of Table 1 is not appropriate. Except for the image block in the first row, the table did not introduce the sources of other image blocks. It is recommended improving Table 1 for better understanding.

11. If there are rotations between blocks, the authors should explain whether the rotations will influence the block selection results.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

16/10/2022

Dear authors,

In the manuscript Efficient and Robust Feature Matching for High-resolution Satellite Stereos you proposes an image block selection method for high-resolution satellite stereos, which processes feature matching in several optimal blocks instead of the entire images. The core of the method is to formulate the block selection into the optimization of an energy function, and a greedy strategy is designed to compute an approximate solution.

General comments

The study is interesting and experimental results indicate the possibility of wider application of the proposed method.

The introduction contains sections that make some claims that are not supported by references.

Conclusion is too short and too generalized. In the Conclusion, you should interpret all the obtained results and present specific claims and confirmations of everything you claim.

Specific comments (are in the manuscript)

-          Line 38-47 - You're making a lot of claims here that you don't provide references for.

-          Line 78-81 - This text does not belong here. You discuss the method in the chapter Method and Materials, and you compare the results in the Conclusion.

-          Line 441 - This is not Table, this is Figure.

-          Line 443 - What does 'Experimental comparisons' mean? Do you mean 'Comparison of experimental results'?

-          Line 502 - Such manuscripts should be written in the third person.

-           

 

Best regards

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have addressed all my concerns.