A Novel Multispectral Line Segment Matching Method Based on Phase Congruency and Multiple Local Homographies
Round 1
Reviewer 1 Report
The revision is acceptable for publication.
Author Response
Thanks for the reviewer's recognition of our work.
Reviewer 2 Report
The manuscript proposes a line segment matching algorithm for multispectral and visible images. The main techniques used are the phase congruency based matching method and multiple local homographies method.
Overall, I suggest the authors to improve the manuscript's organization and writing. For example, the workflow shown in Figure 2 seems to have confused me for many numbers (such as 1, 2, 3, 4) set in the arrowed lines. They could be understood as different steps described in main texts, but they do make the whole structure hard to follow. It also can be found that many formulas are not easy to understand, for example Page 6 Equation (4) “Wo(x, y) represents the weight coefficient, ∆Φm,n(x, y) is the phase deviation”, but the detailed calculations of Wo(x, y) and ∆Φm,n(x, y) are missing. Equation (3), the “atan2” seems to be a function from some algorithm toolboxs, but this is not a conventional expression in an academic publication. There is also a typo in the manuscript, for example, in Figure 9’s caption, “The images in the 1st and 3rd columns belong to LWIR images nad” -> “The images in the 1st and 3rd columns belong to LWIR images and”
Other comments are provides as follows:
1. Page 1, it is said “RGB-RGB based feature matching belongs to single-spectral feature matching”. However, a RGB image actually has three bands corresponding to red, green and blue spectra. Therefore, the above single-spectral statement is questionable.
2. In abstract, it is said “The results shows that the percentage of correct matching (PCM) using PC-MLH can reach 94%, which is significantly outperforms the traditional line segment matching methods with PCMs all lower than 40%.” I would argue whether the benchmark methods are chosen correctly or their parameters are set by a proper way.
3. Figure 9, since the second and the fourth columns are RGB images, it could be better to use the original color picture rather than their gray-level ones.
Author Response
Dear Reviewer,
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 3 Report
The resubmitted manuscript has been substantially revised based on review comments. Various problems and questions contained in the previous manuscript have been resolved. Nevertheless, the gap in accuracy remains in the results of comparative experiments. The accuracy obtained from the two existing methods was less than half that of the proposed method. This is because state-of-the-art methods were not selected for comparison, resulting in large accuracy differences. In the response note, the authors mentioned the difficulty to obtain source codes. Did you contact corresponding authors directly? Did the corresponding authors respond that they could not provide their source codes? Or did you merely search GitHub or other source code posting websites? Public benchmark datasets allow us to compare other methods by extracting the accuracy described in the literature. There is no need to do comparative experiments in your environment. Comparisons can be made in the literature. If not, the reasons should be explained in detail.
Author Response
Dear Reviewer,
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have responded my concerns. But it is suggested that the authors should at least check their manuscript seriously, because typos still can be found after several rounds of revisions apart from its weaker technical contributions. For examples, in Figure 2, “Final Mathes” - > “Final Matches”.
Reviewer 3 Report
The resubmitted manuscript has been appropriately revised.
