Computer-Aided Facial Soft Tissue Reconstruction with Computer Vision: A Modern Approach to Identifying Unknown Individuals
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors-more explanation is needed for: “This means that manual reconstructions are still being used successfully”, p.3, line:101
-a table with the various methods of facial soft tissue reconstruction with their advantages, considering also the digital methods could be helpful for the readers.
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer,
Please see the attachment.
Additionally, I have submitted a version of the revised manuscript with all changes highlighted as a supplementary file to facilitate the review process.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe article presents a detailed exploration of digital facial soft tissue reconstruction techniques and their application in forensic investigations. However, there are several technical aspects that need to be addressed:
1) Despite the use of computer-aided techniques, the interpretation of anatomical soft tissue markers and other biological data still involves a degree of subjectivity. How to overcome the subjectivity was not addressed in the description of the method?
2) The article describes the use of computer vision techniques for face recognition of missing persons. However, these techniques are highly sensitive to minor differences in facial features, lighting conditions, and angles. How to improve the reliability for identification purposes?
3) The threshold value used to determine whether two faces belong to the same individual can significantly impact the results.The default discrimination threshold for face recognition (set at 0.6 in the study) may not be universally applicable. The article does not discuss the potential artifacts or distortions that might arise from threshold selection.
4) While the article highlights the time-saving benefits of digital methods compared to manual reconstructions, the initial setup and processing of digital reconstructions can still be resource-intensive. Key aspects such as the specific parameters used in software applications, the exact algorithms employed, and the preprocessing steps are not fully documented. To ensure reproducibility and credibility, it is essential to provide comprehensive information that allows others to replicate the study.
5) The article uses the Manchester method for both cases, but it does not provide a detailed comparison with other methods. The choice of method can significantly impact the accuracy of the reconstruction, and the article does not justify why the Manchester method was chosen over others.
6) The article presents the results of the reconstructions but does not provide a rigorous validation process. In forensic applications, it is crucial to validate the accuracy of the reconstructions against known benchmarks or datasets. The article does not mention any such validation steps.
7) The article suggests that AI can provide increased objectivity and standardization in facial reconstruction. However, AI models are only as good as the data they are trained on. The article does not address potential biases in the training data or the consistency of the AI models across different datasets.
Author Response
Dear Reviewer,
Please see the attachment.
Additionally, I have submitted a version of the revised manuscript with all changes highlighted as a supplementary file to facilitate the review process.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper outlines the methodology for computer-aided facial tissue reconstruction and identification using face recognition techniques. Comments as follows:
- The abstract is quite long, a-lot of information there should be part of an introduction or background section
- Some paragraphs in the manuscript are too long, containing several different discussions. Hence, these could easily be split into smaller paragraphs for better readability
- L134: "depending on breed and coat colour." - given that the discussion is on a human, this seems our of place or could be better worded
- "data base" - "database" or "dataset"
- Section 3 - "data from imaging techniques" - more information could be given on what this data entails
- The manuscript seems more focused to readers having a biological background, rather than a computing background; hence, depending on where the paper is published, some more background information/explanations on biological terms may be needed
- Section 3.2 L215: "were digitized" - are digitized (keeping the tense consistent with the rest of the discussion)
- Section 3.2.1 - Some more information on the post-processing steps employed would be beneficial
- "Mouth Editor" - is this also an add-on? It should be ensured that any mentions of methods/software used are clearly described and appropriately referenced
- A table to summarise the software, methods, and add-ons for each step of the reconstruction process would be beneficial
- Section 5 L441: "measurement distances as in" - sentence is incomplete
- There are several mentions of "Delib", which should be changed to "Dlib"
- What was the motivation to use Dlib? Were other face detectors and face recognisers apart from those in Dlib considered?
- L483 - "fife" -> five
- More information on the CNN-based face detector and recogniser would be beneficial (i.e. are they based on some existing algorithm or model, such as VGG-Face, etc? Can a reference to the specific methods be provided?)
- Information on the performance of the chosen face detector and face recogniser should be given; otherwise, it is hard to say if the results actually make sense or not (i.e. the methods may not be very good, so the results need to be taken with a pinch of salt)
- The face recogniser was only used on photos of the same person; what are the similarity scores if the pictures of a different person are used? Would they be close to 0.7, higher, lower, ...? This may give a better indication as to whether the face recogniser deems the computer-generated images to be of the same person or of a different person
- The dataset here is very limited, although it is perhaps understandable that it is difficult to acquire skulls and the corresponding true images
- The benefit/usefulness of the face recogniser is not clear here; in real-world scenarios, an image would be compared with numerous other facial images to try and identify the person
- It is perhaps unreasonable to expect a face recogniser trained and tuned on real-world images to perform well when using computer-generated images; this is a branch known as heterogeneous face recognition, where algorithms are constructed to handle face recognition across different modalities (e.g. face photos and face sketches [1]). In cases like this, it might be more useful to rank images according to the similarity score, and determine in which rank the true match lies.
- Section 6.2 - the discussion on AI consistency is not quite true, since different models/methods (trained and tuned on different images and datasets) might yield different results. The advantage is that an individual model would be consistent (but consistency cannot be expected among different models)
In conclusion, the manuscript deals with a very interesting topic, but the suggestions above should be taken into consideration before it is published.
References
[1] C. Galea and R. A. Farrugia, “Matching Software-Generated Sketches to Face Photos with a Very Deep CNN, Morphed Faces, and Transfer Learning,” IEEE Transactions on Information Forensics and Security, vol. 13, no. 6, pp. 1421-1431, Jun. 2018.
The English language is generally very good, but there are some typos that should be fixed and the manuscript could be a-bit better organised, as discussed in the main comments and suggestions.
Author Response
Dear Reviewer,
Please see the attachment.
Additionally, I have submitted a version of the revised manuscript with all changes highlighted as a supplementary file to facilitate the review process.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsAlmost all concerns have been addressed; however, validation remains an issue due to the lack of an available dataset. It is suggested to use FaceForensics++ database for evaluation. For more information, please refer to the following publications:
La Cava， S. M.， Orr， G.， Goldmann， T.， Drahansky， M.， &； Marcialis， G. L. （2023）. 3D Face Reconstruction for Forensic Recognition - A Survey. arXiv:2303.11164v1.
Jain， A. K.， &； Ross， A. （2015）. Bridging the gap: from biometrics to forensics. Philosophical Transactions of the Royal Society B. DOI10.1098/rstb.2014.0254
Tistarelli， M.， Grosso， E.， &； Meuwly， D. （2014）. Biometrics in forensic science: challenges， lessons and new technologies. DOI10.1007/978-3-319-13386-7_12
van Dam， C.， Veldhuis， R.， &； Spreeuwers， L. （2016）. Face reconstruction from image sequences for forensic face comparison. DOI10.1049/iet-bmt.2015.0036
Zeinstra， C. G.， Veldhuis， R. N.， Spreeuwers， L. J.， Ruifrok， A. C.， &； Meuwly， D. （2017）. ForenFace: a unique annotated forensic facial image dataset and toolset. DOI10.1049/iet-bmt.2016.0160
Author Response
Dear Reviewer,
Please see the attachment.
Additionally, I have submitted a version of the revised manuscript with all changes highlighted as a supplementary file to facilitate the review process.
Author Response File: Author Response.pdf
