Next Article in Journal
Statistics and Machine Learning Experiments in Poetry
Next Article in Special Issue
Visual Reconstruction of Ancient Coins Using Cycle-Consistent Generative Adversarial Networks
Previous Article in Journal
Portable XRF Quick-Scan Mapping for Potential Toxic Elements Pollutants in Sustainable Urban Drainage Systems: A Methodological Approach
Previous Article in Special Issue
Making Japenese Ukiyo-e Art 3D in Real-Time
Open AccessArticlePost Publication Peer ReviewVersion 2, Revised

Images of Roman Imperial Denarii: A Curated Data Set for the Evaluation of Computer Vision Algorithms Applied to Ancient Numismatics, and an Overview of Challenges in the Field (Version 2, Revised)

School of Computer Science, University of St. Andrews, St Andrews KY15 5BG, UK
*
Author to whom correspondence should be addressed.
Received: 8 March 2020 / Accepted: 12 March 2020 / Published: 14 June 2020
(This article belongs to the Special Issue Machine Learning and Vision for Cultural Heritage)
Peer review status: 2nd round review Read review reports

Reviewer 1 Alan M. Stahl Princeton University Reviewer 2 Christoph Hartmann Research assistant, Technical University of Munich, Walther-Meissner-Strasse 4, 85748 Garching, Germany Reviewer 3 Eleni Fragaki Leiden University
Version 1
Original
Approved with revisions
Authors' response
Approved with revisions
Authors' response
Approved with revisions
Authors' response
Version 2
Revised
Not approved Approved Approved with revisions
Version 2, Revised
Published: 14 June 2020
DOI: 10.3390/sci2020047
Download Full-text PDF

Version 1, Original
Published: 16 March 2020
DOI: 10.3390/sci2010015
Download Full-text PDF
Automatic ancient Roman coin analysis only recently emerged as a topic of computer science research. Nevertheless, owing to its ever-increasing popularity, the field is already reaching a certain degree of maturity, as witnessed by a substantial publication output in the last decade. At the same time, it is becoming evident that research progress is being limited by a somewhat veering direction of effort and the lack of a coherent framework which facilitates the acquisition and dissemination of robust, repeatable, and rigorous evidence. Thus, in the present article, we seek to address several associated challenges. To start with, (i) we provide a first overview and discussion of different challenges in the field, some of which have been scarcely investigated to date, and others which have hitherto been unrecognized and unaddressed. Secondly, (ii) we introduce the first data set, carefully curated and collected for the purpose of facilitating methodological evaluation of algorithms and, specifically, the effects of coin preservation grades on the performance of automatic methods. Indeed, until now, only one published work at all recognized the need for this kind of analysis, which, to any numismatist, would be a trivially obvious fact. We also discuss a wide range of considerations which had to be taken into account in collecting this corpus, explain our decisions, and describe its content in detail. Briefly, the data set comprises 100 different coin issues, all with multiple examples in Fine, Very Fine, and Extremely Fine conditions, giving a total of over 650 different specimens. These correspond to 44 issuing authorities and span the time period of approximately 300 years (from 27 BC until 244 AD). In summary, the present article should be an invaluable resource to researchers in the field, and we encourage the community to adopt the collected corpus, freely available for research purposes, as a standard evaluation benchmark. View Full-Text
Keywords: coins; review; problems; data corpus; grade; preservation; condition coins; review; problems; data corpus; grade; preservation; condition
Show Figures

Figure 1

MDPI and ACS Style

Arandjelović, O.; Zachariou, M. Images of Roman Imperial Denarii: A Curated Data Set for the Evaluation of Computer Vision Algorithms Applied to Ancient Numismatics, and an Overview of Challenges in the Field. Sci 2020, 2, 47.

Show more citation formats Show less citations formats
Note that from the first issue of 2016, MDPI journals use article numbers instead of page numbers. See further details here.

Article Access Map by Country/Region

1

Reviewer 1

Sent on 01 Apr 2020 by Alan M. Stahl | Approved with revisions
Princeton University

p. 2. "ancient numismatics, i.e., the study of ancient (mostly Roman) currency," - actually, most ancient numismatic research is carried out on Greek coins, whose attributions depend on die and hoard study, unlike Roman imperial coins, whose identification is often obvious from their legends.

2.1 Terminology. This is incorrect use of the word 'type' in a numismatic context. In numismatics, 'type' refers only to the image on the face of the coin. I think the authors are confusing the common meaning of 'type' with the numismatic one, and should use 'issue' for what they term type. This error is repeated throughout the article.

2.3 Die matching. Die study is the basic methodology of most numismatic study, and is vital for discussions of chronology, mint organization, and size of emissions. The authors are incorrect in saying that it is an unexplored challenge in the realm of ancient coin analysis -- in fact it is the aspect that is receiving the most attention from numismatic scholars; a case in point is the long-standing work on automated die identification for the fifth-century tetradrachms of Athens, being undertaken by Dr. Peter Van Alfen of the American Numismatic Society.

2.4 3-D scanning. At Princeton, we have a long-running project on the die analysis of didrachms of Taranto based on 3-D scans performed on our extensive collection, which includes many examples of coins listed as die-linked or die-pairs in the Fischer-Bossert corpus. We have not yet received publishable results due to such problems as registration and the sheer mass of data to be analyzed. We do believe that machine learning is a valuable direction to move in.

3. Degree of wear. We have found that factors other than the degree of wear of specimens pose significant obstacles to machine approaches to coin analysis and identification  - die wear in terms of lowered relief, die wear in terms of spreading and cracking of dies, differences in the centering of dies for striking, differences in the angle of the blow used in striking. And, the reason we decided to work with 3-D scans of coins in our collection, is the vast series of problems introduced by 2-dimensional photography using differential lighting, lenses, and digital transfer to web protocols.

3.1 Denarii. The choice of denarii is sound, but attention should be paid to recent work by Butcher and Ponting showing the degree to which the composition of denarii in the Roman Empire varied considerably over time.

Table 1. This should be presented in chronological rather than alphabetical order.

4. I fail to see any explanation of what degree or area of analytical knowledge is gained by the consideration of the degree of preservation of studied examples in the application of computer-assisted numismatic study. Basically this article introduces one of many factors to be taken into account in machine-assisted numismatic analysis but doesn't explain its role in such analysis or give examples of results obtained through it.

Response to Reviewer 1

Sent on 14 Jul 2020 by Ognjen Arandjelovic

We are genuinely thankful to the reviewer for the comments - we found them all helpful and constructive, so thank you!

Having reread out statement, we now see that we did not express ourselves clearly enough, which cased confusion where there is actually no disagreement at all. The "mostly Roman" was not meant to "ancient numismatics" (as we fully agree with the reviewer's comments) but "application [of computer vision] to ancient numismatics". To text in the parentheses was removed to avoid confusion.

We fully agree with the reviewer that we were overly lax and colloquial with out terminology here and appreciate this being pointed out. Hence the correction has been made.

We are referring to die study using computer vision and machine learning, not die study by humans i.e. numismatists, which as the reviewer states is major area of research. This discrepancy, if you will, is precisely why we highlight this challenge as a worthwhile one in the context of computer vision and machine learning.

We fully agree which is precisely why we felt that there was a need for a corpus which would allow this to be studied systematically by researchers in computer vision.

As per our response above (and as stated in the manuscript), the main purpose of collecting this corpus was not to study the degree of wear but rather to provide a benchmark for the study of computer vision algorithms and in particular their sensitivity to wear - something that has generally not been done well in the existing literature.

Reviewer 2

Sent on 07 Jun 2020 by Christoph Hartmann | Approved with revisions
Research assistant, Technical University of Munich, Walther-Meissner-Strasse 4, 85748 Garching, Germany

The Paper is well written and tackles an interesting target. My only siggestion is mabey to include also a connection to the field of experimental numismatics (see for example Hartmann, C.; Hammerl, F.; Volk, W.: Experimental analysis of Roman coin minting. Journal of Archaeological Science: Reports 25, 2019, 498-506), since the proposed methods would be also beneficial for researchers in this field.

Response to Reviewer 2

Sent on 14 Jul 2020 by Ognjen Arandjelovic

We are genuinely thankful to the reviewer for the positive comments as well as for drawing our attention to a paper which we agree is of interest. Thank you! The amendment has now been done.

Reviewer 3

Sent on 04 Jun 2020 by Eleni Fragaki | Approved with revisions
Leiden University

The topic of this paper is very interesting and the suggested research has definitely a strong potential. However, both the content and the presentation should be revised in order to reach the aims of the study.

Even if there are no previous attempts to provide an overview of the applications of machine learning in the field of numismatics, the related literature needs to be summarized and evaluated in the introduction.

The enumeration and discussion of the challenges in this field has to take more appropriately into account the concerns of specialists in ancient currencies and the exact meaning of the terminology must be reexamined and respected. This part of the article should rather focus on technical aspects and describe the specific tools and methods that can be used to address these questions.

The idea to collect a curated set of coins in various stated of preservation has an obvious advantage from an archaeological point of of view. However, the exact technical implications of this choice could be more extensively analyzed. If the readership is supposed to include experts in a numismatics and hobby collectors the technological parts must be formulated in a more accessible and explicative way. Even if the text is intended only for computer scientists, the parts dealing with numismatics should be reconsidered.

Response to Reviewer 3

Sent on 14 Jul 2020 by Ognjen Arandjelovic

We are thankful to the reviewer for the kind comments as well as all the helpful suggestions.

We fully appreciate why the reviewer feels that way but we hope that some genuine disagreement will be permitted here. In particular, there are a number of papers which provide such overviews, and which we cite, so we would consider yet another review as such to be superfluous, unnecessarily adding redundant content, and taking away attention from the main contributions of the work. As we said, we hope that the reviewer will allow us, the authors, the prerogative of choice in this balancing act. Numerous references are provided for the interested reader.

Where there is previous work we have indeed referenced a representative sample of articles. For problem areas where more work is due, the aim of our article is to draw the attention of the computer vision community interested in this realm of research (but few, if any, of whom are experts in ancient numismatics itself) to them. When and if we develop concrete solutions we will publish them, rather than speculate a priori.

The nature of the special issue and indeed the present article certainly do not make the target audience neither expert numismatists nor lay hobby collectors, but rather the section of the computer vision community interested in applying their expertize in the field of ancient numismatics.

Reviewer 4

Sent on 01 Jun 2020 by Thomas Faucher | Approved with revisions
Chargé de Recherche classe normale, IRAMAT-CRP2A (CNRS, UMR 5060), Bordeaux

The article presents the latest considerations on the computer vision developments on the field of numismatics.

I personally doubt the interest of such a dataset which is obviously directed towards the commercial side of numismatics. The databases of major collections (British Museum London, BnF Paris, Münzkabinett Berlin, American Numismatic Society New York), as portals on specific coinages (RPC online, RIC online, Pella, Seleucid coins online, etc…), have florished in the past years presenting large datasets available with good quality images and curated datasets. The classification by wear is mostly used for commercial purposes for obvious financial matters even though some scholars have been trying to use them (See J.-M. Doyen, L'indice d'usure des monnaies en tant que substitut aux indices de fréquence, in S. Krmnicek & J. Chameroy (eds), Money Matters, Coin finds and ancient coin use, Bonn, 2019, p. 21-30).

The distorted version of the work on computer analysis focused on Roman numismatics shown in this article mostly forget that the majority of the work done on computer die recognition have been made on Greek and Celtic coins. Die studies have been massively used in Greek numismatics because datasets are more manageable than the Roman ones for which corpora of most issues make them impossible to study with a die analysis.

As the article is entitled “overview of challenges in the field”, it would be therefore appreciated to enlarge the references for the different aspects of Ancient Numismatics.

p. 2: ancient numismatics cannot be considered as “mostly Roman”

p. 2: “An important consideration in numismatics regards the condition of a particular coin”. This is true only for the commercial part of the numismatics, not for the research part. I do not see what this sort of consideration “the price of the coin, and thus its affordability as well as its investment potential, are greatly affected: The cost of the same type can vary by 1–2 orders of magnitude” help the debate in the article.

p. 6:” type matching as probably the most important and pervasive problem”. It is true for a commercial use. The interest of academic relies more certainly on die matches.

p. 7: “To the best of our knowledge, die matching remains an entirely unexplored challenge in the realm of automated ancient coin analysis”. This is suprising to read. Efforts have been made since the 1970’s to be able to perform automatic die studies. Once again, the limited view of the article on Ancient Roman numismatics eliminates most of the research held in Greek and Celtic numismatics, which are more advanced than Roman numismatics on Die analysis. The bibliography would be too long to provide here. For the latest : Taylor, Z. M., and van Alfen, P. Coins and computation: New developments in the computer-aided die study. https://www.youtube.com/watch?v=wF6aZdhc0wg, 2020. ANS Money Talks: Numismatics Conversations ; van Alfen, P. The computer aided die studies program. http://numismatics.org/pocketchange/cads/, 2017. Accessed: April 27, 2020 ; Sofiane Horache, François Goulette, Jean-Emmanuel Deschaud, Thierry Lejars, Katherine Gruel. AUTOMATIC CLUSTERING OF CELTIC COINS BASED ON 3D POINT CLOUD PATTERN ANALYSIS. 2020. ⟨hal-02559952⟩.

p. 9: “the difficulty of obtaining a curated data set; hence, our present effort and contribution.” Databases of major collections are now online. Curators of each institutions have supervised the information in these databases. Nowadays, hundreds of thousands of coins are available through these institutions and even reunited in portals, offering major curated datasets.

Response to Reviewer 4

Sent on 14 Jul 2020 by Ognjen Arandjelovic

As we trust our comments below will illustrate, we would at least somewhat disagree with the first claim (and there is also something to be said about being rather dismissive of the "commercial" side of numismatics; also see below). There is no disagreement here as regards the reviewer's second claim, but it is rather beside the point. As stated in the manuscript, the main purpose of collecting this corpus was to provide a benchmark for the study of computer vision algorithms and in particular their sensitivity to wear - something that has generally not been done well in the existing literature. This kind of information (coin condition) is generally not available in academic data sets, and is certainly not curated in the same structured way as our data set which comprises the same issues across different conditions.

Having reread out statement, we now see that we did not express ourselves clearly enough, which cased confusion where there is actually no disagreement at all. The "mostly Roman" was not meant to "ancient numismatics" (as we fully agree with the reviewer's comments) but "application [of computer vision] to ancient numismatics". To text in the parentheses was removed to avoid confusion.

Not exclusively so. Consider for example the Portable Antiquities Scheme in the UK (https://finds.org.uk/database/search/results) which records the general public's finds. Huge numbers of coins need to be identified so that their location and other relevant find information can be recorded. Automating this process would certainly be invaluable. Understanding how different automatic, computer vision based methods are affected by wear is extremely important in developing a working solution that would help such community initiatives aimed at helping scholars.

While the reviewer may not like that (and we fully appreciate this view!), a huge number of ancient coins are owned by private individuals, are sold and traded, etc., and an individual parting with a coin for a fraction of its price because it was misidentified by an automatic algorithm is something that most hobby collectors would care about.

The key words here are "aided" and "automatic" - there is a major difference. In the former, and indeed in the articles highlighted by the reviewer, a whole load of manual intervention, work, and effort is needed. This is by no means "automatic", at least not in the sense that a computer vision researcher would use the term.

Indeed but not with the type of information in our data set. As stated in the manuscript, the main purpose of collecting this corpus was to provide a benchmark for the study of computer vision algorithms and in particular their sensitivity to wear - something that has generally not been done well in the existing literature.

Back to TopTop