The Potential for Image Analysis in Numismatics
The Potential for Image Analysis in Numismatics
Abstract and Keywords
The systematic study of coinage at the level of individually engraved die revolutionized numismatics; however the laborious nature of such work has severely limited its application. Die studies are important for attribution and chronology as well as for quantification. Exhaustive study on coinage reveals great evidences particularly in economic history. This chapter discusses the potential posed by image analysis and photography on the field of numismatics. It discusses the principal technique challenges, with a view to stimulating discussion as to the best way forward.
The application of new imaging techniques to numismatics has not progressed beyond the use of digital images in connection with databases, and for web and conventional publication. Such applications are of real importance but lie well within the scope of what can be done already, and hence are not of interest here. It is unlikely that image enhancement will be of great significance, except, perhaps, in microscopic metallurgy, because numismatists can normally see and reproduce what is necessary for the purposes of study. Coins were mass-produced: if one specimen is hard to read, it will usually be more useful to locate a better specimen than to enhance the image of the poor one. Where there is spectacular scope for progress is in image comparison, but no real start has been made in this area. This paper is therefore about significant potential, rather than current developments.
The systematic study of coinage at the level of the individually engraved die (alongside the scientific study of coin hoards) revolutionised numismatics, but the laborious nature of such work has severely limited its application.1 Die studies are crucial not only for attribution and chronology, but also for quantification. There remains a great deal of evidence to be exploited, particularly for economic history. This paper sets out the potential for image analysis in this area, and also the principal technical challenges, with a view to stimulating discussion about the best way forward. The technical difficulties are significant, but the problems are interesting and solutions may have wider applications.
The development of the methodology of the die study and its application from the 1870s for mint attribution were themselves linked to a major revolution in imaging, namely the advent of photography.2 It is a thesis of this paper that the new revolution in imaging has the potential to facilitate another quantum leap in the subject.
The Die Study
The die study is a simple but very powerful method. Most coins in antiquity were manufactured by striking a disk of metal (‘blank’ or ‘flan’) between two hand-engraved punches (‘dies’) (Figure 11.1). Few actual dies survive, but it is possible to gain a reasonably full knowledge of the original dies through visual comparison of the coins struck from them. Individual dies may well have struck 20,000 or 30,000 coins, and it is normal for each known die to be represented by a number of surviving coins. Statistical techniques allow us to estimate the original number of dies from the degree of duplication among the surviving examples. Quantification is to some degree controversial, not least because the range of the number of coins struck from a die can only be estimated from comparative studies and experimental archaeology. Even without this difficulty, calculations often produce results with wide margins of error. But no mint records survive from the ancient world, and the original number of dies is our best guide to the absolute size of an issue of coins.3 The topic is important both for financial history (since one of the main reasons for striking coin sin antiquity was to enable states to make payments) and for economic history (where the quantity of coins in circulation informs our models of money use).4
The lower die (‘obverse’), protected in an anvil, and the upper die (‘reverse’), in the form of a hand-held punch, wore out at different rates, and were replaced as necessary. Thus it is possible to identify sequences of dies linked together by studying the coins struck from them (Figure 11.2). Coins produced from a linked sequence may with reasonable certainty be attributed to broadly the same context (mint and date), so that the die study marked a radical improvement on earlier subjective attributions by style alone.
Our ability to utilise die studies for attribution and quantification has been limited by the sheer quantity of comparisons that have to be made for large issues of coinage. In principle, within each issue (defined for this purpose as a group of coins with the same images and inscriptions, and of the same denomination), each coin must be compared with every other coin. In practice, large issues may be broken down into smaller groups by detailed considerations (for example, the spacing of the inscription) so that there is no possibility of die links between groups (and hence no need to make comparisons). But the procedure remains a daunting prospect for very large coinages. Thus the coinages of Athens and Rome at the heights of their respective empires were so huge that few substantial die-studies have been attempted.5 To give an idea of scale only, the principal coinages in gold and silver in circulation in the Hellenistic world around 290 BC have been estimated to have been struck from 65,000 drachma-equivalent dies,6 and the Roman Republican mint is believed to have used up to 2400 dies for silver denarii in a single year.7 It is perhaps symptomatic that the two largest die-studies of particular coinages under the Roman empire remain unpublished.8 In general the application of the methodology has been patchy: die studies have often been undertaken where needed for attribution, but rarely for quantification alone.
Crawford made a pioneering attempt to quantify the coinage of the Roman Republic by extrapolating from a few small die-studies, using the relative frequency of issues in hoards as an index of their relative size.9 Duncan-Jones applied a similar technique to the imperial coinage of the Roman principate.10 Such attempts are to be applauded, but introduce further and very significant margins of error. Moreover, although hoards can give an idea of relative size, die studies provide the only significant approach to absolute quantification. No one would seriously doubt that it would be better to have both full die-studies and indices of frequency in hoards, so that these two approaches to quantification, each with their own problems, may be used to control each other.
The scale of the material presents a problem for taxonomy and attribution, as well as for quantification. The ‘Berlin Corpus’, the initiative of Mommsen and the great Swiss numismatist Imhoof-Blumer to create a classification of all Greek coins, has suffered from slow progress following the ‘scientifically correct’ but burdensome decision to work at the level of the die (the crucial debate took place in the 1890s and early 1900s).11 Thus the sheer quantity of comparisons to be made in die studies is an obstacle to the advancement of the subject in a number of fundamental areas. It is here that computerised comparison of digital images has great potential.
Such a project will not be easy because the images to be compared have been subject to many distortions, at the time of striking (I), through subsequent wear and damage (II), and at the time of image-capture (III). If computerised comparison is to be effective, it will be necessary to find a way of identifying and discounting such distortions.
Distortions at Time of Striking
The fact that coins were struck by hand introduces a range of distortions in the images to be compared.
1 The penetration and angle of strike varies. The human eye (at least) can cope with this easily in most cases, but some instances may be very misleading. Modern experimental striking has produced unexpected variations between products of the same die (Figure 11.3).12 The problem is at its most extreme in those archaic coinages which were struck on one side with an ‘incuse punch’ (Figure 11.4). The degree to which the punch was driven into the metal, and the angle of entry, produced enormous variations in the shape of the indentation on the coin (marks within the punch mark may be a better guide to die identity). (p.111) The only consolation is that die studies of such material are by far the hardest to perform in all ancient numismatics. The positioning of the image on the flan varies, and it is not uncommon for coins to be struck sufficiently off centre for part of the image to be missing from the coin altogether (Figure 11.5).
2 The flan may be struck more than once by the die and the two impressions may not be precisely aligned, resulting in a ghosting (Figure 11.9 below) or plainly disjointed effect (Figure 11.6)—‘double struck’.
3 The die itself may be damaged in the course of use, and the ‘die break’ may be progressive (Figure 11.7). The damage to the die shows up as an additional feature in relief on the coin (so that a crack in the die results in a ridge on the coin). This is not all bad news: the numismatist knows that two similar coins with consistent die-breaks are virtually certain to be from the same die. Presumably a computer could also be trained to use this short cut. Die breaks may also be useful in providing the chronological direction of a die sequence: a coin with a particular die-break must have been struck later than a coin from the same die without the die break.
4 A further complication, which may also be beneficial, is that, sometimes, existing coins were used as flans (resulting in an ‘overstrike’). The image on the original coin may show through and thus ‘distort’ the superimposed image (Figure 11.8). Such distortions must be discounted for the purpose of die comparison, but overstrikes also provide vital information for the numismatist in the form of a cast iron stratigraphy. The coin used as a flan must be earlier than (p.112) the coin struck over it. It can be extremely difficult to recognize the under type from the traces left, and computerised imaging might facilitate this by ‘lifting’ one image off the other, perhaps in much the same way that two superimposed texts may be separated.
Once in circulation individual specimens may be damaged in a variety of ways. They may be holed (Figure 11.9) or cut (Figure 11.10), or deliberately stamped with punches bearing images and/or inscriptions (‘countermarks’) (Figure 11.11). They may also be corroded (either adding to, or subtracting from, the surface), and surface colour may be varied. Surface discoloration may be a significant problem for coins known only from directly taken images(photographic or digital), which is why numismatists have often preferred to work from plaster casts. But by far the most significant of these problems will be mundane coin wear—almost all coins are worn to some degree (Figure 11.10). Some-how it will be necessary to model wear, possibly drawing on the packages used to model landscape erosion.
The coins may be a vailable to be imaged under fixed conditions, but they are in many locations, and the numismatist also has to cope with plaster casts, conventional photographs, and digital images. Perhaps 40 per cent of the material is known only from photographs in trade sale catalogues, and increasingly from commercial digital images on the web. So there can be no ducking problems arising out of the variability of image capture.
Shadows are an obvious problem for 2D images which have been taken under variable lighting conditions. The direction, nature, intensity, and number of the light source(s) may all vary. Controlled stereo imaging or 3D imaging might provide an answer for new images, but cannot help with the substantial body of existing images. Presumably similar problems of shadow are addressed in the analysis of aerial and satellite photographs.
To make an impact we need not only a suitable technology for image capture, but also a powerful facility for automated image comparison. This need not be perfect: it would still be helpful to be given all certain and probable die-links (the level of probability needs consideration). One could then settle for working within defined margins of error (for the purposes of quantification), or resolve ‘difficult cases’ by visual inspection where possible. Even so, a major challenge will be to define key variables, and to model forms and levels of variability displayed by products of the same die, and to differentiate that from variability between products of different dies.
It is possible to work at the level of only 2D images: the human eye and brain can perform a perfectly satisfactory die-study from photographs alone. Here fingerprint technology might have interesting lessons to teach, but in numismatics no progress has been made beyond visual comparison of images superimposed on film or screen.
However, to reduce 3D objects (albeit with shallow relief) to 2D may be to throw away valuable information which could help with some of the distortions outlined above. For example, one might model a die from a surviving coin and then fit it to another coin. This might help to resolve problems resulting from angle of striking and degree of penetration, and from striking off-centre. It might also provide an approach to coin wear. Since wear can only detract from relief, and not add to it, a‘die’fitted (p.113) to a coin from that die cannot overlap it (the wear would show up as a gap between ‘die’ and coin). 3D imaging in itself removes problems of surface discoloration. For the comparison of 3D images in general it may be fruitful to draw on technology from medical imaging.
Much material is known only from photographs, so that any successful technique will eventually have to be able to compare 2D and 3D images. There may be advantages here: for example, a 3D model might allow one to model lighting variability in 2D images. Perhaps the availability of some evidence in 3D will allow the 2D images to be reconstructed in 3D. Since the relationship of 2D to 3Dimages has been the subject of much exciting work described in this volume, it is apparent that there is scope for interesting work in this area.
From the numismatic point of view it is easy to imagine that different classes of coin may be better approached by different techniques. Thus it may be simpler to study large coins with relief on both sides in 2D, but smaller coins with incuse punch marks in 3D. Whatever the case, it is certain that a portfolio of techniques will be required. A further practical concern is that, since coins often cannot be removed from museums, imaging equipment will need to be portable and relatively inexpensive. Technical problems will be demanding, but a start might be made by testing techniques on a small well-defined sequence of coins which display a range of image distortions.
One interesting consequence of such a project would be that automated die-studies could provide a control on those performed solely by human judgement. It is generally believed by numismatists that, although die studies (like most things) may be done badly, two adept practitioners will replicate each other’s results. It would be interesting to test this by automated comparison, which will require the judgemental processes to be made explicit. It might also be helpful to museums to know whether 3D imaging has significant advantages over2D imaging, before they invest in large-scale projects to image their collections. More than that, there is at least the potential for new imaging techniques to make possible die-studies on a scale not practical up to now, and thus to transform a subject.
I am grateful to Alan Bowman, without whose prompting this paper would not have been written, and to Volker Heuchert, Henry Kim, and Greg Parker for useful discussion and practical help.
Beer Tobey, L. and Tobey, A. G. (1993), ‘Experiments to simulate ancient Greek coins’, in M. M. Archibald and M. R. Cowell, Metallurgy in Numismatics, vol. 3 (London): 28–35.
Bland, R. F. (1991), ‘The coinage of Gordian III from the mints of Antioch and Caesarea’, unpublished doctoral thesis (Institute of Archaeology, University College, London).
Callataÿ, F. de. (1995), ‘Calculating ancient coin production: seeking a balance’, Numismatic Chronicle 155: 289–311.
Callataÿ, F. de. (1997), Recueil quantitatif des émissions monétaires hellénistiques (Wetteren).
Callataÿ, F. de., Depeyrot, G. and Villaronga, L. (1993), L’argent monnayé d’Alexandre le Grand à Auguste (Bruxelles).
Crawford, M. H. (1974), Roman Republican Coinage (Cambridge).
Crawford, M. H. (1983), ‘Numismatics’, in M. H. Crawford (ed.), Sources for Ancient History (Cambridge): 185–233.
Crawford, M. H. (1990), ‘From Borghesi to Mommsen: the creation of an exact science’, in M. H. Crawford, C. R. Ligota and J. B. Trapp (eds.), Medals and Coins from Budé to Mommsen (London): 125–32.
Doyen, J.-M. (1989), ‘L’atelier de Milan (258–268): recherches sur la chronologie et la politique monétaire des empereurs Valérien et Gallien’, unpublished doctoral thesis (Université Catholique de Louvain).
Duncan-Jones, R. (1994), Money and Government in the Roman Empire (Cambridge).
Göbl, R. (1978), Antike Numismatik (Munich).
Howgego, C. (1990), ‘Why did ancient states strike coins?’, Numismatic Chronicle 150: 1–25.
Howgego, C. (1992), ‘The supply and use of money in the Roman world 200 bc to ad 300’, Journal of Roman Studies 82: 1–31.
Howgego, C. (1995), Ancient History from Coins (London).
Kaenel, H.-M. von (1991), ‘“…ein wohl grossartiges, aber ausführbares Unternehmen”: Theodor Mommsen, Friedrich Imhoof-Blumer und das Corpus Nummorum’, Klio 73: 304–14.
Kraay, C. M. (1956), The Aes Coinage of Galba (New York).
Kraay, C. M. (1976), Archaic and Classical Greek Coins (London).
MøRkholm, O. (1982), ‘A history of the study of Greek numismatics. III. c. 1870–1940: the scientific organisation’, Nordisk Numismatisk Årsskrift: 7–26.
MøRkholm, O. (1991), Early Hellenistic Coinage: From the Accession of Alexander to the Peace of Apamea (336–188 BC) (Cambridge).
Volk, T. R. (1987), ‘Mint output and coin hoards’, in G. Depeyrot, T. Hackens and G. Moucharte (eds.), Rythmes de la production monétaire, de l’antiquité à nos jours (Louvain-la-Neuve): 141–221.
Walker, D. R. (1988), Roman Coins from the Sacred Spring at Bath (Oxford).
(4) For examples of the use of quantification to address financial history and money use respectively, see Crawford (1974); Walker (1988). For the broad issues and an analysis of the problems involved, see Howgego (1990, 1992).
(6) The estimate of actual number of dies was lower: the use of ‘drachma-equivalent dies’ is a convention to express comparative value. For example, a tetradrachm (four drachma) coin die is counted as four drachma-equivalent dies. See de Callataÿ et al. (1993), 43, for 290 BC, and annexe I (p. 97) for later issues; de Callataÿ (1997).