Target Image Mask Correction Based on Skeleton Divergence

Round 1

Reviewer 1 Report

In the paper, a method for target image mask correction is presented. The description of the approach and the reported experimental evaluation are superficial and require substantial changes. The paper contains a lot of unsupported claims. The detailed comments are as follows:
1. The obtained reduction of the impact of cumulative results is not shown.
2. Details of used techniques, such as bilateral filter, hierarchical clustering on others, are not shown. Their usage is not justified experimentally. What were other choices? What made them suitable for the task? What would happen if other methods were used here instead?
3. Page 3, line 133-144. Please also explain `v’.
4. The reduction of the calculations is neither discussed nor shown (mentioned in line 185).
5. The improvement of the efficiency of the system is not investigated (mentioned in line 186).
6. The influence of the parameters of the method on its performance is not shown. Some of the parameters are introduced and not mentioned again (like initial threshold – line 190).
7. The advantage of the hole filling is not validated experimentally.
8. The optimization task is not (formally) defined. Also, constraints are not discussed.
9. The used dataset is not characterized.
10. How do we know that the presented masking effects are correct (Fig. 1-2)? Efficiency and accuracy (line 235) cannot be seen here. Similar observations can be made for findings reported in Section 3.1. It seems that the mask correction is presented using two frames (Fig 1 and Fig 2-3). Are they similar in other cases?
11. The advantages over other methods are not discussed.
12. Some sentences contain repetitions (lines 35-36, page 1; lines 141-143, page 3). Please rewrite them.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposed an image mask correction based on the skeleton
 divergence. There are some issues, which should be addressed:

It is hard to understand the overall process of this paper, please design an image to describe the overall process of the proposed model (including input, internal processes, output) The sensitivity of the presented methodology to the set-up of its parameters should be investigated Most of the references are outdated, please consider updating them accordingly. The conclusion appears to be just a summary of work done, rather than a critical appraisal of the presented research work. When the proposed algorithm outperforms others, why is this the case? Any insights what makes this algorithm accurate, from a mathematical/algorithmic point of view? The dataset must be described in more details (How many images were used, size, format,....). There should be more description of the development environment, such as programming language, os, parameters of the algorithm. So the interest readers can replicate the work. The computational complexity (time, computing power) of the proposed model must be examined

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

In the opinion of the reviewer, the revised paper is in a better shape of presentation. However, the following issues should be taken into account in the revision:
1. Please enlist the contributions of this study at the end of Section 1. Please highlight the presented novelty there.
2. The performance of the method with other clustering algorithms should be shown. Also, it should be clarified how the distance between clusters was measured and why.
3. It is written that “the directional corner can reflect the trend and degree of curvature of the global edge”. The high-curvature regions require an additional explanation, considering the size and shape of such a region. In which circumstances such a region starts being considered “a high-curvature”?
4. Please correct the caption of Fig. 2. It should start with the capital letter.
5. Please highlight the best results in Tables 1 and 3 (boldface).
6. Please add information on the camera used to capture the depth and color sequences while introducing the TUM RGBD dataset. Was it Kinect?
7. It is written that TOPSIS “orders ideal merge strategies”. What does it mean? Why ideal? Please rewrite this.
8. There are missing arrows and shifted texts in Fig. 1. Also, the arrow that comes from the space between “Simplify” and “Association” blocks is somewhat confusing.
9. The sourcecode of the method should be made publicly available. Please ensure the repeatability of the main results. If the size of the dataset prevents it, the results for single frames shown in figures should be easy to replicate by a reader.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I sincerely thank the authors for answering my questions. However, there are more issues that need to be addressed in the latest version of the manuscript:

Figure 1. the algorithm flow is shown in the graph, but the explanation for each step in the subsections (2.1=>2.4) did not follow the flow in Figure 1. Therefore, it is very difficult for the reader to follow. Equation 2 line 144, what is the extract directed rotation angle of the two adjacent edges vectors being used in this paper. The authors just explained about its minimum and maximum range. Similarly for equation (3), line 182 and equation (4) line 202. The value for these parameters must be described. Section 3.3 and 3.4 title is similar. Table 1, add a citation for each method. Table 2, How did the authors compute the computing time. Is it for one image or a dataset?. Why  I think it is suitable to combine table 1 and table 1 into one table. Table 3, what is prop is it the proposed method, Please write the full word.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

All comments have been addressed, so I propose this paper for publication by the algorithm journal.