Two New Ways of Documenting Miniature Incisions Using a Combination of Image-Based Modelling and Reflectance Transformation Imaging

Round 1

Reviewer 1 Report

(The editorial office has changed file type as pdf, please find at attachment.)

Please see attached txt file

 

If you don't find a 162 line text file attached, please let me know and I will email it to you

% wc remotesensing-797531-peer-review-v1.txt 

     162    1408    9001 remotesensing-797531-peer-review-v1.txt

Comments for author File: Comments.pdf

Author Response

 

Dear reviewer,

 

Thank you for the kind review and for the suggestions! The article is not only intended for experts in this very limited field, so your comments are most welcome.

 

First, regarding your general comments:

1. This is, of course, not up to us, but would be very interesting.
2. We have tried to acquire data for a fair comparison of methods and software, and have included a lot of details in the appendixes for others to inspect. In the latest version we have also included Ground Sample Distance (GSD) values to Table 4. These show that there are some differences in the quality of the source images. This is acknowledged and discussed in the Discussion chapter, but does not impact the main conclusions of the article. The example you mention - where one software were able to align pictures but the other were not - could have been fixed by taking more pictures. But as one of the software packages DID manage to align these pictures, this was left as it was since it showcased a difference in the programs ability to align.
3. We have tried to make the tables more clear by expanding the captions to them, without repeating too much of the main text there. We have also taken away some unnecessary information, and divided Table 1 into two tables. The tables are also made more clear by one of the recorded surfaces being removed from the article, as it was not really adding anything useful to it.

 

We will answer the more specific comments below.

• Multiple paragraphs in the abstract is not allowed, but the slight adjustments you suggested have been added.
• Line 42: 2.5D is defined in section 2.2.2., we have added a reference to this.
• Line 56: The keyword NEW has been added to the title, and to the abstract.
• Line 99-102: This is a misunderstanding, MS and RC do not implement RTI methods. The line says, "all 3D models were made or attempted made using two  3D processing software packages." As mentioned earlier, RTI do not create 3D models.
• Line 107-108: Yes
• Line 129-130: No. The line says "RTI is a method of capturing and enhancing information on a surface." There are other methods, as the one you mention.
• Line 137-138: DEM and DTM does not belong to RTI. "To separate these (meaning the RTI 2.5D models) from other types of 2.5D models, such as (..), they (meaning the RTI 2.5D models) will be referred to here as RTI models."
• Line 138-141: DTM and DEM was not explicated, this has been corrected. The reference to Polynomial Texture Maps (PTMs) has been removed, while we view the rest of the acronyms here as important and all are now explicated.
• Line 157: "ø" is sometimes used for diameter. As many probably do not know this we are now using just "diameter" instead.
• Line 167-174: Yes, the difference between the two combination methods is in the data processing. The images for both combination methods were acquired by photographing the surfaces with the RTI dome taking 50 pictures from each 28 (Surface A) and 25 (Surface B) different positions. The 50 pictures from each position had 50 different illuminations. For Combo 1, these images were post-processed using IBM software – even though IBM software is usually used for post-processing images with even illumination and where every image is from a unique position. For Combo 2, each set of 50 pictures were “distilled” into one RTI model first. Then, two JPG exports from each of these positions were exported from the RTI software and post-processed with the IBM software (the reason for exporting two pictures for each position instead of one is described in the paragraph below in the article). We have added this extended description of the process to the article and hope this makes the process more clear.
• Line 184: This was mentioned above. One of the IBM software packages were able to align the images. This highlighted this program’s apparent superior (at least, in this case) ability to align, and we did not see the need to go back to take more photographs.
• Line 273: The descriptions in the captions of the tables and figures have been improved.
• Line 292-294: It says “Evaluating the accuracy of these models is (..) not adequate when based on quantifiable variables, because IBM generates 3D models, and RTI generates 2.5D models”. And later: “Since the RTI models do not contain vertices, assessment (corrected from acquisition) of accuracy also needs to be carried out visually.” Hence, the following comparison between the 3D models and the RTI models is done visually.
• Line 306: In the article we use them as synonyms. As it says: “(..) the accuracy of these models, ie. the level of detail (..).”
• Line 310-311 and Figure 1: No longer relevant, since the original Surface B and this figure have been removed from the article. But yes, we would say so.
• Line 312-332: Hopefully, we have managed to do both.
• Line 326: This has been added.
• Figure 2: This is a problem that stems from the inability to choose illumination angle on the Normals Visualization in RTIViewer. Ideally, the angle of the light chosen for the 3D models should have been the same as these, but instead the angle that seemed to showcase the most details of the 3D models were chosen.
• Figure 3: The caption has been improved. It should also be noted that we made a mistake when uploading the article online – a version where the images were compressed were used instead of the original, which had full resolution images. In the current version the full resolution images have been replaced. In addition, a selection of images from the two figures with the most pictures have been removed to be able to show larger versions of the remaining images. This will make them somewhat easier to see, though the best solution is to zoom into them, either in a digital version of the article or in the images uploaded online.
• Figure 4: The RTI models are the most accurate, though the 3D models are more detailed than they seem when represented as compressed small-version images (see above).
• Figure 5: Same as above. The 3D models are not failed, but the RTI models do exhibit more details.
• Figure 6: The missing captions have been added.
• Table 4: The table has been improved, we think, by adding the evaluations you suggested for another table in the Conclusion chapter.
• Lines 393-396: Yes, the Combo 1 method do seem to be more robust to noise/data acquisition errors. This has been pointed out, and the term “rebuked” replaced with “dismissed”, which better fits our intention.
• Line 426 – end: This was a good suggestion, which we chose to incorporate instead in Table 4.

Best regards,

Dag-Øyvind Solem and Erich Nau, NIKU

Reviewer 2 Report

This work describes an innovative method to documenting incisions, that uses digital photogrammetry and reflectance transformation imaging integrated together.

The paper is well written, the research well presented. The topic is interesting, albeit a niche one.

Regarding the presented research, I have only one general observation: for the surface B a real integration between the two techniques was not realized (no Combo are presented), therefore the presentation of this case, in my opinion, unnecessarily weighs down the work, without giving a real contribution in the evaluation of the added value of using them together. I would eliminate the surface B, limiting the treatment to A and C; this would also make Table 1 more readable.

Some little specific comments:

• In paragraph 2.1 I would add some images of the analysed surfaces; I saw that they are in figure 8,

but I would anticipate them here

• Line 20: I would replace the semicolon with the colon
• Line 21: …there is currently non documentation technique that is optimal regarding…maybe was no documented technique
• Line 110: follow, not follows
• Line 112: better homologous than similar

Author Response

 

Dear reviewer,

 

Thank you for the kind review and the suggestions! You are right, the original Surface B was no longer relevant for the article and has been removed. The article was originally mainly intended as a RTI/IBM comparison review, but as the focus of the article shifted towards the new combination methods, Surface B became redundant without us noticing. Some additional steps have been taken to make the tables more clear, including splitting Table 1 into two tables and removing the picture per minute ratios.

 

Regarding your other comments;

• Two figures of the recorded surfaces have been added to paragraph 2.1
• Line 20: The semi-colon has been replaced.
• Line 21: "Documentation technique" has been changed to "recording technique".
• Line 110: Misspelling corrected
• "Similar" has been replaced by "homologous".

 

Best regards,

         Dag-Øyvind Solem and Erich Nau, NIKU

Reviewer 3 Report

The topic is interesting and well presented. The case studies are well illustrated and the bibliography is updated. Appendices are really useful for better understanding equipments (hardware and software), settings, data sets and workflow.

Honestly, the only observation concerns the missing information on the resolution (precision and accuracy) of the final results (in terms of GSD, deviation, etc.). The number of the vertices is not so relevant since in the software the user can select how dense the point cloud should be.

I suggest to add few lines about these issues; a table could better show these data.

Author Response

 

Dear reviewer,

 

Thank you for the kind review and suggestions! The GSD values have been added to Table 4. At some point we were thinking about making more in-depth comparison figures in CloudCompare of the full-resolution model details that eventually ended up in Blender. The plan was to use the Combo 2 models as "blueprints" to compare the other models, as they visually were deemed to be the most accurate. Ultimately, learning to do this precisely was not prioritized but we hope to be able to do this in our later work.

 

Regarding your comments about the vertice count, we had completely missed that MetaShape allows a custom Face Count for the meshes that are higher than the default High setting. This is now explained in the article, thank you for pointing that out. It will be interesting to see how this will affect future comparisons.

 

Best regards,

           Dag-Øyvind Solem and Erich Nau, NIKU Norway

Reviewer 4 Report

 This work is related to enlarge the opportunity that offers Digital 3D documentation methods such as Image-Based Modelling (IBM) and laser scanning towards the documentation of objects that needs do define very small details as miniature incisions such as graffiti. The work presents ways to combine IBM and RTI and evaluate these different techniques with the aim of comparing accuracy and efficiency in relation to implementation times and costs. It was also examined the influence of two different 3D processing software packages on these factors: the widely used MetaShape (MS) and a more expensive option, RealityCapture (RC).

The article shows that there is currently no optimal documentation technique for all the considered factors and that the methodologies s that combines RTI and IBM seems to be the best way to record miniature engravings as in the case analyzed. In addition to cost efficiency, the main difference between the 3D software packages tested is that RC is much faster than MS. There were no significant differences in the assessment of reliability.

The work presents a good clarity and organization of the contents which demonstrate an adequate scientific deepening with accurate bibliographic references. I suggest to the authors to have a look of the works of Bologna group of research (F. I. Apollonio, P. Clini, M. Gaiani The third dimension of Leonardo’s Vitruvian Man, 2019) about Leonardo Drawings.

Author Response

 

Dear reviewer,

 

Thank you for the kind review and your suggestion for litterature!

We have not been able to get a hold of the suggested article yet, as it is for some reason not available for purchase on Amazon.com from Norway. We have, however, sent a request for the article to the authors via Researchgate, and look forward to read it.

 

Best regards,

Dag-Øyvind Solem and Erich Nau, NIKU, Norway

Reviewer 5 Report

Over the years I have reviewed many manuscripts focused on different applications of remote sensing and high-resolution recording in archaeology. These have become extremely important as these technologies have become common across all types of research and heritage management. I say all this to put the following statement in context: this is one of the best reviews of a high-resolution recording method I have ever read. The authors have gone to lengths to explain the reasoning behind their tests of techniques (IBM, RTI, and IBM+RTI combined), provided several clearly described tests of technique (Surfaces A-C), describe data collection with rare clarity and completeness, outline the specifications of post-processing workflow (software, costs, time), provide standard and novel (combined) results that are rigorously compared, disseminated their raw data and digital products (via Figshare), discuss the results including several possible confounding factors that might have been impacting the results. There is little else one could ask for. I highly recommend it for publication.

Some very minor suggestions. The paper is well written. I would however suggest replacing the sexist term “manmade” with some other term (human made, or artificial) and “For years there has been projects that” should be “have.” Table 1, text could be cleaned up. Table 2, and elsewhere, the $ sign goes before the number ($4000 rather than 4000$).

One suggestion that the authors may consider, but should not be required, is evaluating the results in the way remote sensing does as true/false positive/negative (i.e., treating them like they were cropmarks). As in, if you compared it on a pixel by pixel based with some traditional technique, how would they compare? It would also be interesting to see these results transformed to a vector format to better compare them in terms of some basic metrics (length, width). My last suggestion, again it should not be required for publication, would be to think about replacing “real” with something more technically specific, like “current condition.” Real could be misinterpreted. Also, perhaps the other quality that they are discussing could be termed legibility, especially since some of the images are text.

Author Response

 

Dear reviewer,

 

Thank you for your kind review and your suggestions!

At one stage of the process, the plan was to make visual comparisons of full-resolution parts of the 3D models using a software like CloudCompare to get the kind of more pronounced visual representation of the differences in scale you suggest. In the end, learning how to do this properly was not prioritized as it did not include a way to compare the RTI data. However, we hope to be able to do this type of more detailed comparisons in our further work.

 

You, and other reviewers, have commented in essence that the tables should be clearer, so we have tried to do this - both by simplifying the tables themselves and also by describing them further in the captions.In regard to the other details you point out; "manmade" is replaced with "human made", some typos have been found and corrected, and the dollar signs have been switched around. The term "real" has been replaced with "3D model accuracy", which is more in line with our intentions. The albeit vague term "accuracy" has been kept, since we - based on other feedback - chose to remove the surface with actual letters on (since it was not recorded with the combination techniques, it wasn't really important for the article).

 

Best regards,

Dag-Øyvind Solem and Erich Nau, NIKU, Norway