A Method for Regularizing Buildings through Combining Skeleton Lines and Minkowski Addition
Round 1
Reviewer 1 Report
This manuscript proposed a new approach that combines skeleton lines and the Minkowski addition for regularizing building outlines. The topic is very interesting, however, some drawbacks need to be addressed. The major comments of the study are:
First, the authors did a relatively detailed literature review in the Introduction. However, the reviewer thinks this section's logic should be revised (it should be easy for readers to follow). The authors should highlight their main contributions or novelty.
Second, the authors should report which methods were used in Q-GIS and why this method was used. Also, the result and discussion parts seem too thin.
Third, there is no legend for Fig.11. what do red boxes, red lines, and green boxes stand for?
Fourth, the authors mentioned that they used high-resolution remote sensing images. What is the spatial resolution? When were the images published?
Minor editing of English language required
Author Response
Thanks a lot for your kind comments, We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
Comment1: First, the authors did a relatively detailed literature review in the Introduction. However, the reviewer thinks this section's logic should be revised (it should be easy for readers to follow). The authors should highlight their main contributions or novelty.
Response1: We are sorry that this part was not explained clearly in the original manuscript. We have modified the Introduction section to make it easier to read.
Comment2: the authors should report which methods were used in Q-GIS and why this method was used. Also, the result and discussion parts seem too thin.
Response2: We have explained this in section 3.2 of the manuscript. QGIS is used as development tool in this study, For the simplification of polygon features, QGIS provides the corresponding interface simplify(double tolerance) in the QgsGeometry class, which will return a simplified polygon feature using a specified tolerance value.
We have modified the results of the experiment, we divided the study object into seven types according to the shape of the buildings, choosing five buildings for each type to be analyzed and explained in the discussion section.
Comment3: Third, there is no legend for Fig.11. what do red boxes, red lines, and green boxes stand for?
Response3: We are sorry that this part was not explained clearly in the original manuscript. We have added legends to Figure 11.
Comment4: the authors mentioned that they used high-resolution remote sensing images. What is the spatial resolution? When were the images published?
Response4: We are sorry that this part was not explained clearly in the original manuscript. We have explained it in section 3.1 of the manuscript. The spatial resolution of the high-resolution remote sensing image is 0.67m, published in 2020.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors discuss a method developed for reconstructing the footprint of buildings from RGB image. The authors' intention is clear; however, the authors should make a comparison with real data, e.g. obtained on a sample of buildings. This way, the proposed approach can be treated statistically, i.e. OA etc. Therefore, I suggest taking into account and adding in the paper some works present in the scientific literature:
https://doi.org/10.3390/ijgi10100697
https://doi.org/10.1080/22797254.2017.1416676
https://doi.org/10.1016/j.isprsjprs.2020.07.011
https://doi.org/10.1109/TGRS.2005.843569
Minor revision
specify GIS acronym
introduce in fig.10 a scale bar, north arrow and frame of the reference.
I suggest improving figure 13 by excluding the formulas
Author Response
Thanks a lot for your kind comments, We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
Comment1: The authors discuss a method developed for reconstructing the footprint of buildings from RGB image. The authors' intention is clear; however, the authors should make a comparison with real data, e.g. obtained on a sample of buildings. This way, the proposed approach can be treated statistically, i.e. OA etc. Therefore, I suggest taking into account and adding in the paper some works present in the scientific literature:
https://doi.org/10.3390/ijgi10100697
https://doi.org/10.1080/22797254.2017.1416676
https://doi.org/10.1016/j.isprsjprs.2020.07.011
https://doi.org/10.1109/TGRS.2005.843569
Minor revision
specify GIS acronym
Response1: Thank you so much for your comment, especially the references listed helped me a lot! We have made detailed revisions in the Introduction, Experiment section of the manuscript. In addition, we have added acronyms to the manuscript where the GIS first appears.
Comment2: introduce in fig.10 a scale bar, north arrow and frame of the reference.
Response2: We have carefully considered the suggestions and made some changes in Fig. 10. In order to better display the simplification results, by constructing the mapping relationship of vector raster data, the mapping of vector data into the high-resolution remote sensing image is more responsive to the effect of simplification. The detailed revision results in the manuscript are shown in Figure 10.
Comment3: I suggest improving figure 13 by excluding the formulas
Response3: Thank you so much for your comment, we have removed and reformatted the formulas in Figure 13.
Author Response File: Author Response.docx
Reviewer 3 Report
The article is well structured and of interest to the readers. Figure 13 and the equations shown below it need to be reformatted. I would make some mention of the possibility of using deep learning for building classification already prepared within multiple GIS software. To this end, I would ask you to see the following articles from which to compare your algorithm with those of artificial intelligence
Yuan, J. (2017). Learning building extraction in aerial scenes with convolutional networks. IEEE transactions on pattern analysis and machine intelligence, 40(11), 2793-2798.
Tripodi, S., Duan, L., Poujade, V., Trastour, F., Bauchet, J. P., Laurore, L., & Tarabalka, Y. (2020, September). Operational pipeline for large-scale 3D reconstruction of buildings from satellite images. In IGARSS 2020-2020 IEEE International Geoscience and Remote Sensing Symposium (pp. 445-448). IEEE.
Pepe, M., Costantino, D., Alfio, V. S., Vozza, G., & Cartellino, E. (2021). A novel method based on deep learning, GIS and geomatics software for building a 3D city model from VHR satellite stereo imagery. ISPRS International Journal of Geo-Information, 10(10), 697.
Author Response
Thanks a lot for your kind comments. We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
Comment: The article is well structured and of interest to the readers. Figure 13 and the equations shown below it need to be reformatted. I would make some mention of the possibility of using deep learning for building classification already prepared within multiple GIS software. To this end, I would ask you to see the following articles from which to compare your algorithm with those of artificial intelligence
Yuan, J. (2017). Learning building extraction in aerial scenes with convolutional networks. IEEE transactions on pattern analysis and machine intelligence, 40(11), 2793-2798.
Tripodi, S., Duan, L., Poujade, V., Trastour, F., Bauchet, J. P., Laurore, L., & Tarabalka, Y. (2020, September). Operational pipeline for large-scale 3D reconstruction of buildings from satellite images. In IGARSS 2020-2020 IEEE International Geoscience and Remote Sensing Symposium (pp. 445-448). IEEE.
Pepe, M., Costantino, D., Alfio, V. S., Vozza, G., & Cartellino, E. (2021). A novel method based on deep learning, GIS and geomatics software for building a 3D city model from VHR satellite stereo imagery. ISPRS International Journal of Geo-Information, 10(10), 697.
Response: Thank you so much for your comment, especially the references listed helped me a lot! First, we have reformatted the equation in Figure 13. Second, we have attempted to use the open source QGIS software to build a deep learning method to do the classification of buildings, but the classification results were not satisfactory. However, we are working on using artificial intelligence techniques to simplify and merge buildings. In addition, we will attempt to use multiple GIS software comparisons in the future study.
Author Response File: Author Response.docx
Reviewer 4 Report
This paper introduces a method that combines skeleton lines and Minkowski addition for building regularization. While the overall work is interesting, the authors did not clearly define the problem or highlight their contribution. Additionally, there are issues with the presentation of the method and experiments. It is recommended that the paper undergo major revisions before being reviewed again.
1) The literature review needs substantial improvement. On the second page, the paper simply lists studies conducted by different authors without providing necessary summaries or conclusions. Additionally, on the third page, there is a redundant and ineffective statement, "where the experimental results showed...". Furthermore, there is a noteworthy concentration of references from Chinese authors, with a significant portion of the references being sourced from Chinese journals.
2) The paper is hard to follow because it primarily consists of descriptions of technical details. It is not clear which techniques are borrowed from others and which ones are the author's own innovations. Additionally, it is recommended to place the flowchart at the beginning of the paper to provide readers with an overall understanding of the study.
It is unclear what the exact innovation of the paper is—whether it is the achievement of adaptive adjustment of size and orientation of structure elements or the resolution of the conflict problem.
3) What kind of problem is "Building Regularizing" mentioned in the paper, and what are its inputs and outputs? It is important to provide a formal definition to ensure clarity, as not all readers can be expected to be experts in this field.
4) The paper introduces Minkowski addition but fails to explain how it is utilized in the subsequent methodology section.
5) The paper mentions real-time updating of the data, but it does not provide a comparison of the efficiency between the proposed method and traditional methods. The paper mentions that the proposed method is slower than QGIS, it does not specify the extent of the performance difference. This raises concerns about whether the efficiency of the proposed method could potentially pose a bottleneck during its implementation.
6) What method does QGIS use? It is crucial to provide a clear and detailed explanation of the method employed by QGIS. Additionally, it is important to determine whether the method used by QGIS is considered the current optimal solution for addressing the building regularization problem at hand.
7) The outcome of building regularization in the top-left corner of Figure 11f seems to be less satisfactory compared to Figure 11e. The building in Figure 11f loses too many intricate details during the regularization process, in contrast to Figure 11e.
8) Was only one building object selected for each quantitative statistic in Table 1? To ensure the robustness of the results, it is advisable to select multiple samples from the study area for calculation.
9) What is the reference object for calculating the "change in area" and "surface distance"? Additionally, in Table 1, there are cases where the “change in area" is 1, suggesting that s1 and s2 are completely equal?
10) the discussion part is missing.
Other minor comments:
1) what's the meaning of the sentence “Based on an existing method proposed earlier by this author” in abstract？
2) The caption for Figure 2 is not on the same page as the figure itself.
3) At the top of page 10, it describes the steps, but why does it also mention the results?
4) In section 3.2, when describing the spatial extent, are latitude and longitude used? Why are there negative signs?
no comments
Author Response
We are very grateful to Reviewer for reviewing the paper so carefully. We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
Comment1: The literature review needs substantial improvement. On the second page, the paper simply lists studies conducted by different authors without providing necessary summaries or conclusions. Additionally, on the third page, there is a redundant and ineffective statement, "where the experimental results showed...". Furthermore, there is a noteworthy concentration of references from Chinese authors, with a significant portion of the references being sourced from Chinese journals.
Response1: We have made revisions in the literature review section to provide a summary of the cited literature, and the references have been adjusted.
Comment1: The paper is hard to follow because it primarily consists of descriptions of technical details. It is not clear which techniques are borrowed from others and which ones are the author's own innovations. Additionally, it is recommended to place the flowchart at the beginning of the paper to provide readers with an overall understanding of the study.
It is unclear what the exact innovation of the paper is—whether it is the achievement of adaptive adjustment of size and orientation of structure elements or the resolution of the conflict problem.
Response2: We am sorry that this part was not clear in the original manuscript. The purpose of this research is to propose a new method for building regularization, traditional morphological methods usually keep the direction and size of the structuring elements unchanged during processing, the method proposed in this paper produces conflicts during processing due to the need to rotate the structuring elements, which affects the quality of the results, in order to solve the above problems, this paper proposes a method based on the angular bisector is used to solve the conflicts and to improve the effectiveness of this paper's method.
In addition, we put the flowchart at the beginning of the paper to make it easier for the reader to understand it.
Comment3: What kind of problem is "Building Regularizing" mentioned in the paper, and what are its inputs and outputs? It is important to provide a formal definition to ensure clarity, as not all readers can be expected to be experts in this field.
Response3: We am sorry that this part was not clear in the original manuscript. There is always a significant loss of relevant building cues due to occlusion, poor contrast, shadows, and disadvantageous image perspective. Under the influences of these factors, the extraction of building outlines often becomes jagged and noisy, making it difficult to apply directly to cartographic generalization. We explain inputs and outputs in detail in section 2.2 of the paper.
Comment4: The paper introduces Minkowski addition but fails to explain how it is utilized in the subsequent methodology section.
Response4: We am sorry that this part was not clear in the original manuscript, since the original manuscript flowchart is located later in the article, makes it difficult for the reader to have an overall understanding of the methodology.
The Minkowski addition algorithm requires two regions, one of which is the region to be processed, which in this paper corresponds to the skeleton line of the building, and the other is the structuring element, which needs to be customized in terms of its shape, size and orientation. A detailed explanation is given in section 2.2 of the paper.
Comment5: The paper mentions real-time updating of the data, but it does not provide a comparison of the efficiency between the proposed method and traditional methods. The paper mentions that the proposed method is slower than QGIS, it does not specify the extent of the performance difference. This raises concerns about whether the efficiency of the proposed method could potentially pose a bottleneck during its implementation.
Response5: The method in this paper is slower than QGIS in terms of efficiency, and we are currently working on further optimization of this method to improve the efficiency of the program.
Comment6: What method does QGIS use? It is crucial to provide a clear and detailed explanation of the method employed by QGIS. Additionally, it is important to determine whether the method used by QGIS is considered the current optimal solution for addressing the building regularization problem at hand.
Response6: QGIS is used as development tool in this study, For the simplification of polygon features, QGIS provides the corresponding interface simplify(double tolerance) in the QgsGeometry class, which will return a simplified polygon feature using a specified tolerance value. Thus, it was chosen to be used as a comparison of the method in this study. Detailed explanation has been added to the paper in section 3.2.
Comment7: The outcome of building regularization in the top-left corner of Figure 11f seems to be less satisfactory compared to Figure 11e. The building in Figure 11f loses too many intricate details during the regularization process, in contrast to Figure 11e.
Response7: We have rechecked the result of the top-left corner of Figure 11e and Figure 11f. When simplifying the skeleton lines, due to the different scales of the buildings, in order to ensure the overall optimality of the result of the skeleton line simplification, there is a possibility that some of the building features are lost, thus affecting the quality of the regularization.
Comment8: Was only one building object selected for each quantitative statistic in Table 1? To ensure the robustness of the results, it is advisable to select multiple samples from the study area for calculation.
Response8: Thank you so much for such great suggestions! In order to evaluate the building regularization method proposed in this paper, seven types (C-shaped, F-shaped, H-shaped, I-shaped, L-shaped, T-shaped, V-shaped) of buildings from the study area were selected for analysis, and select five typically distributed buildings for each type. The details are explained in section 3.2 of the paper.
Comment9: What is the reference object for calculating the "change in area" and "surface distance"? Additionally, in Table 1, there are cases where the “change in area" is 1, suggesting that s1 and s2 are completely equal?
Response9: The reference object for calculating the "change in area" and "surface distance" is the building before simplified.
Comment10: the discussion part is missing.
Response10: We have added a discussion in section 3.2 of the paper.
Comment11: what's the meaning of the sentence “Based on an existing method proposed earlier by this author” in abstract？
Response11: We are very sorry for the confusion caused by the lack of clarity in the paper. Previously we have done a preprocessing work with the extraction and simplification of building skeleton lines. The sentence “Based on an existing method proposed earlier by this author” in abstract corresponds to our previous work.
Comment12: The caption for Figure 2 is not on the same page as the figure itself.
Response12: Thank you so much for pointing out the details in the paper. We have made changes in the paper.
Comment13: At the top of page 10, it describes the steps, but why does it also mention the results?
Response13: We are very sorry for the confusion caused by the lack of clarity in the paper. We adjusted the logic of the article presentation.
Comment14: In section 3.2, when describing the spatial extent, are latitude and longitude used? Why are there negative signs?
Response14: We are very sorry for the confusion caused by the lack of clarity in the paper. We use latitude and longitude in the article to indicate the extent of the study area. It has been explained in the article.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
No more comments
Author Response
We are very grateful to Reviewer for reviewing the paper so carefully. We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
Author Response File: Author Response.docx
Reviewer 2 Report
The authors made the changes to the paper taking into consideration all my suggestions and comments. Therefore, the paper can be published in the present form.
Author Response
We are very grateful to Reviewer for reviewing the paper so carefully. We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
Author Response File: Author Response.docx
Reviewer 4 Report
1) There are still some noticeable errors, such as the format error for reference [18] and the missing numbering for the reference "Tripodi et al." In the response, they use "we am" for several times.
2) There are still quite a few Chinese references, such as [4], [7], [9], [15], [16], [23]. Please verify if the authors have English versions of these papers published, or whether these references are necessary to include.
3) I appreciate that the authors provided a summary in the literature review section. However, I couldn't discern the logical connections among these references. It seems like the authors listed some references without explaining the purpose behind including them.
4) the discussion part is still missing. Please note that the discussion should be a separate section in the paper.
5) The authors have pointed out that their method is slower than QGIS. However, it would be beneficial to provide a detailed efficiency comparison to quantify the extent of this slowness and to offer insights into potential strategies for optimization. This information could be appropriately included in the discussion section of the paper.
6) While the authors explained the method used by QGIS, the question of "whether the method employed by QGIS is considered the current optimal solution for addressing the building regularization problem at hand" still remains unresolved.
7) The issue with Figure 11f needs further clarification in the paper. It should be explained under what circumstances this problem arises and how much impact it has on the algorithm's accuracy, which could also be included in the discussion part.
8) Regarding the experiments in Figure 12 and Table 1, what criteria were used for selecting these 5 samples? Were they chosen randomly? Why not use all the data for this experiment?
9) When describing geographical coordinates, the authors are still using negative numbers, and they omitted the degree symbol (°).
no comments
Author Response
We are very grateful to Reviewer for reviewing the paper so carefully. We have carefully considered the suggestions and made some changes. The revised details are highlighted in red.
1) There are still some noticeable errors, such as the format error for reference [18] and the missing numbering for the reference "Tripodi et al." In the response, they use "we am" for several times.
Response1：The format error reference and the missing numbering for the reference have been addressed in the paper.
2) There are still quite a few Chinese references, such as [4], [7], [9], [15], [16], [23]. Please verify if the authors have English versions of these papers published, or whether these references are necessary to include.
Response2：These Chinese references have been deleted from the paper.
3) I appreciate that the authors provided a summary in the literature review section. However, I couldn't discern the logical connections among these references. It seems like the authors listed some references without explaining the purpose behind including them.
Response3：The literature review section provides a further summary of the current research methods, details were presented in the Introduction section.
4) the discussion part is still missing. Please note that the discussion should be a separate section in the paper.
Response4: The discussion section has been added to Part 4 in this paper.
5) The authors have pointed out that their method is slower than QGIS. However, it would be beneficial to provide a detailed efficiency comparison to quantify the extent of this slowness and to offer insights into potential strategies for optimization. This information could be appropriately included in the discussion section of the paper.
Response5: The detailed efficiency comparison has been added to the Table 1 in this paper. The causes were analyzed and optimization strategies were presented in this paper.
6) While the authors explained the method used by QGIS, the question of "whether the method employed by QGIS is considered the current optimal solution for addressing the building regularization problem at hand" still remains unresolved.
Response6: It is difficult to define which method is optimal. As described in the literature(A hybrid approach to building simplification with an evaluator from a backpropagation neural network), “it is almost impossible to appropriately simplify all buildings in a dataset by the standalone application of existing algorithms”. The method employed by QGIS is more stable in simplifying complex buildings。
7) The issue with Figure 11f needs further clarification in the paper. It should be explained under what circumstances this problem arises and how much impact it has on the algorithm's accuracy, which could also be included in the discussion part.
Response7: To address this issue, we have explained in detail in the paper, with details as shown in Figure 14.
8) Regarding the experiments in Figure 12 and Table 1, what criteria were used for selecting these 5 samples? Were they chosen randomly? Why not use all the data for this experiment?
Response8: Firstly, buildings have a distinct orthogonal character compared to other natural geographic features. These seven building types are the most widely distributed in the study area. Secondly, these samples were randomly selected. Thirdly, for example, there are rings inside some, leading to significant deviations in the results. Further processing is needed. Therefore, we randomly selected buildings with typical distribution in the study area as samples.
9) When describing geographical coordinates, the authors are still using negative numbers, and they omitted the degree symbol (°).
Response9: We've modified it in the paper.
Author Response File: Author Response.docx
