3D capture, representation and manipulation of cuneiform tablets
Sandra I. Woolley *, Nicholas J. Flowers , Theodoros N. Arvanitis , Alasdair Livingstone ,
Tom R. Davis and John Ellison
The University of Birmingham, Edgbaston, Birmingham, England
Harvard University, Cambridge, MA 02138, USA
This paper presents the digital imaging results of a collaborative research project working toward the generation of an on-
line interactive digital image database of signs from ancient cuneiform tablets. An important aim of this project is the
application of forensic analysis to the cuneiform symbols to identify scribal hands.
Cuneiform tablets are amongst the earliest records of written communication, and could be considered as one of the original
information technologies; an accessible, portable and robust medium for communication across distance and time. The
earliest examples are up to 5,000 years old, and the writing technique remained in use for some 3,000 years. Unfortunately,
only a small fraction of these tablets can be made available for display in museums and much important academic work has
yet to be performed on the very large numbers of tablets to which there is necessarily restricted access.
Our paper will describe the challenges encountered in the 2D image capture of a sample set of tablets held in the British
Museum, explaining the motivation for attempting 3D imaging and the results of initial experiments scanning the smaller,
more densely inscribed cuneiform tablets. We will also discuss the tractability of 3D digital capture, representation and
manipulation, and investigate the requirements for scaleable data compression and transmission methods. Additional
information can be found on the project website: www.cuneiform.net
Keywords: 3D imaging, cuneiform tablets, laser scanning, forensic analysis, data archival, data compression.
1. CUNEIFORM TABLETS
Cuneiform tablets1 are rectangular slabs of sun-dried or fired clay, moulded by hand and inscribed when wet with a wedge-
shaped stylus. Cuneiform (from the Latin word cuneus, meaning wedge) was the most common system of writing in
Western Asia from 3000 BC through to the middle of the first millennium BC when it was gradually replaced with various
alphabetic systems of writing. The signs are typically 3 mm in height, with about 30 signs per line and 25 lines of writing
both on the front and back of the tablet.
The Scottish engineer James Nasmyth first came to live in London in 1829 and discovered the British Museum and their
rapidly growing collection of cuneiform tablets, on which the work of decipherment was progressing apace. He later wrote
in his biography 2,
"I was specially impressed and interested with the so-called "Arrow-head" or "Cuneiform Inscriptions" in the Assyrian
Department. ... I was particularly impressed with the precision and simple beauty of these cuneiform inscriptions".
This interest motivated him to investigate the mechanics of cuneiform production including the reed stylus and the roller
seal, now commonly referred to as a cylinder seal, as shown in Figure 1.
The dimensions of a cuneiform tablet can range from those of a credit card to those of a palmtop computer, the so-called
"monster tablets", and were inscribed by means of a square-profiled reed stylus that made triangular indentations, or
wedges, with “tails”.
* Correspondence: Email: S.I.Woolley@bham.ac.uk; WWW: http://www.eee.bham.ac.uk/woolleysi; Telephone: +44 121-414-7521;
FAX: +44 121-414-4291
Three-Dimensional Image Capture and Applications IV, Brian D. Corner, Joseph H. Nurre, Roy P. Pargas,
Editors, Proceedings of SPIE Vol. 4298 (2001) © 2001 SPIE. · 0277-786X/01/$15.00 103
Figure 1: James Nasmyth the 19th century engineer and his early descriptions of cuneiform production 2
An example of a neo-Babylonian tablet and script details are shown in Figure 2. As can be seen, each sign consists of a
number of wedges. Along with ancient Egyptian and Chinese writings, these tablets are among the earliest records of
written communication. Cuneiform writing itself developed a number of different varieties, and the tablets were also later
inscribed with early forms of the alphabet that we still use today. The earliest examples are up to 5,000 years old, and the
writing technique remained in use for some 3,000 years. The writing was deciphered in the last 150 years; hundreds of
thousands of tablets have been discovered and are now preserved in museum archives. Unfortunately, only a small fraction
of these tablets can be made available for display and much important academic work has yet to be performed on the very
large numbers of tablets to which there is necessarily restricted access.
Figure 2: Examples of cuneiform tablets a) neo-Babylonian b) detail of Neo-Babylonian script from tablet (a)
and c) neo-Assyrian detail
Proc. SPIE Vol. 4298
2. THE CUNEIFORM DIGITAL FORENSIC PROJECT (CDFP)
The CDFP3 began in January 1999. It is a multidisciplinary project funded by The University of Birmingham, U.K. Our
team includes an Assyriologist, a forensic handwriting expert and engineers specialising in database design and digital
imaging; all are members of the University of Birmingham, UK.
The main objective of the CDFP is the development of a digital cuneiform sign database4 requiring digitization of cuneiform
tablets and signs, and the scientific forensic analysis of cuneiform sign configurations.
Much of the cuneiform material is fragmentary; excavations performed in the 19th century lacked the rigor of modern
scientific archaeology, with the result that much material is unprovenanced. Traditional cuneiform sign lists, both ancient
and modern, have always recognised the existence of a variety of script forms pertaining to different areas and periods in the
long history of the diverse civilisations that used the cuneiform script. In addition, scholars who have specialised in
particular cultural complexes - rural, urban, temple, palace - with large numbers of tablets belonging to one area and period
have frequently maintained that they can recognise the handwriting of individual scribes. The digitization of the cuneiform
texts is an important first step in the development of a forensic database for scholarly research.
It is hoped that our research activities will assist in enabling tablets to be assigned scientifically to the archives or contexts
to which they belong, with implications for the documentary history of the areas concerned, thus providing objective
support for expert knowledge. The digital imaging aspects of the work involve 2D and 3D imaging, and the delivery of a
CD and www-accessible digital library for both research and teaching activities.
3. DATASETS FROM THE 1ST MILLENNIUM BC
Two culturally specific areas have been targeted for concentrated work by the CDFP, an area where kings and courtesans
rub shoulders with an elite of scholars on one hand, and a rather plebian ancient college of further education on the other.
Each of these areas provides an extremely complex set of data.
The first of these brings the project to the project to the palaces and royal halls of the mighty Assyrian Empire in the 7th
century BC. It is known that a group of twenty or so learned men, the "inner circle", served as advisers to the kings
Esarhaddon (680-669 BC) and Ashurbanipal (668-627 BC), while other groups of scribes and scholars at the court are also
represented by copious cuneiform documentation in the form of letters and reports. The project has been using digital and
forensic analysis to study the structure of the system of advisory personnel of the royal court. This investigation recently led
to the discovery that a tablet found at Nineveh may in fact have been written personally by or for Ashurbanipal, probably
while he was crown prince. It is well known that Ashurbanipal claimed literacy in a royal inscription and that as king he
founded at Nineveh a massive library, some twenty-three thousand tablets and fragments of which are now in the British
Museum. In addition, several crudely written tablets seem to be the work of Ashurbanipal himself, since the script shows
similarities to the script of tablets signed by the scribe Balasi, who was appointed as Ashurbanipal's teacher. A letter to
Ashurbanipal from his young brother (laku) survives. Perhaps most fascinating is a letter to Ashurbanipal's young wife
Libbi-ali-sharrat ("O-city-she-is-queen!") from his elder sister Sherua-etirat ("Goddess-Sherua-is-the-one-who-saves!)
concerning writing a tablet and reading it out. Sherua-etirat reminds Libbi-ali-sharrat that she is a "daughter of Akkad" and
now a member of the ruling family in the "house of succession". By establishing a database of sign forms and individual
cuneiform script types (hands) the project is contributing to the investigation of the cult literacy and the indigenous interest
in cuneiform surrounding Ashurbanipal as crown prince of Assyria.
The second area involves an archive of tablets which has recently been literally pieced together by I.L.Finkel of the Dept of
Western Asiatic Antiquities of the British Museum. These tablets date to the reigns of the Persian kings of Babylon Darius
and Xerxes, about two hundred years later than the material referred to above. Certain peculiarities of the documents such
as spelling mistakes and multiple copies of the same documents in different hands show that what is involved here is a
scribal school. The schoolmaster himself can be identified and it seems that he used various kinds of written material at his
disposal in his instruction. The level is higher than that of a primary school and could be compared to a college. There are
two varieties of instructional material, namely documents of a business nature such as contracts, and medical texts. This
Proc. SPIE Vol. 4298 105
was then a Babylonian college of business and medicine. It provides an ideal dataset for the project aims of forensic
handwriting analysis and database construction.
The digital imaging of cuneiform tablets poses a number of challenges. Central to the challenges is the fact that many of the
tablets contain densely compacted inscriptions on both sides, with text running over rounded edges and frequently
continuing along each side edge, making the complete 2D capture of symbol sets problematic even with many exposures
from different angles.
Our initial experiments involved SLR photography at the British Museum, using a four-light bed with adjustable camera
height and with tablets mounted on a sand tray to stabilise their orientation (Figure 3). Lighting proved problematic, not
least because of repeated burns from the directional spotlights. The depth of the impressions also made the selection of
correct lighting both difficult and time-consuming.
North-west lighting is usually preferred for illumination of cuneiform script and did produce the best results. Soft, neon,
north-west lighting was later adopted; however, many images, and occasionally whole reels of film were discarded on
Analogue photographic capture was ruled out as a tractable method for capture of the cuneiform datasets primarily because
of the difficulty of preserving scale and context information, the inability to annotate (i.e. properly label) images, the
continuous and complex adjustments to accommodate the tablet surfaces, and, importantly, the absence of direct visual
4.1. Image resolution and format
In addition to cuneiform experts, others may also wish to view cuneiform tablets. For example, the optical recognition of
characters presents a number of interesting challenges to researchers5, 6. Some viewers may only interested in the general
shape and size of the tablets, and a simple, low-resolution model may suffice. Other scholars may be interested in reading
the symbols and would therefore require more intelligible, higher resolution renditions. However, the highest level of detail
is required by forensic analysts. Discussions with forensic experts indicate that a resolution of approximately 0.025mm is
required to distinguish between different scribal hands.
Our interest in both sign recognition and forensic analysis dictated high-resolution capture, which proved particularly
difficult in the case of the smaller tablets. Conventional NTSC video cameras are convenient to use but only provide 480
lines of resolution, although many video capture devices offer higher resolution. This increased resolution is obtained by
interpolation and smoothing, and although these techniques improve the visual quality of the image, the additional
resolution is artificial and cannot be used for forensic analysis.
The lossless 256-color GIF image format was selected as the most efficient format for 2D images because the high-
resolution color information of the baseline JPEG standard was not required. GIF was also preferred because of its
universal support, in particular, its support in www browsers.
New still digital cameras now provide significantly higher resolutions but have practical limitations, in particular, storage
capacity and battery life. We used two digital image capture techniques; a simple video camera capture system and a 3.34
megapixel digital camera (The Nikon CoolPix 990), using both zip and CD disc media. The video camera capture
resolution proved insufficient for forensic purposes. The digital camera produced images of much more acceptable
resolutions, but both techniques required continuous lighting adjustments and frustrating power supply, laptop computer,
software and media logistics.
Other logistical problems included the regular transport of various pieces of equipment to the museum (much improved by
the museum's kind co-operation in storing the light bed and camera), the precious nature of the objects which required
Proc. SPIE Vol. 4298
extreme care in handling at all times, the lack of on-site high-bandwidth networking, and, the need to continuously annotate
captured details (requiring our expert Assyriologist).
When reading the tablets, experts tend to rotate them constantly; grossly in order to present all the signs to inspection, but
also subtly, in order to use light and shadow to bring out the indentations clearly. In a digitised tablet, this manipulation
would ideally be enabled by high-resolution 3D rendering, rotation control and light source adjustment.
Figure 3 : Image capture in the British Museum cuneiform archives
5. 3D SCANNING
There are a number of different techniques for capturing 3D models, each offering different scanning areas and resolutions.
The Digital Michelangelo project at Stanford University7 involved scanning larger objects, e.g., Michael Angelo's statue of
David, at a resolution of up to 0.25mm using laser triangulation rangefinding. The 3D -Matic project8 at the Turing Institute
and The University of Glasgow used ‘photogrammetry’ to give a resolution of 0.5mm. Commercial laser stripe triangulation
systems9 can deliver resolutions of 0.1mm (potentially 0.05mm). Surface contact probes can achieve similar resolutions,
but they make physical contact with the object obviating their application to precious or fragile objects and the scans are
very slow; for example, it would take approximately three days to scan one face of a 50mm by 70mm cuneiform tablet. The
best technique may be moiré, which theoretically offers very high resolution10 although commercially available portable
moiré scanners offer resolutions of closer to 0.2mm. Table 1 summarizes the differences between the various scanning
Technique Scan size
(mm) Scan time for one side
(s) Equipment Cost
Photogrammetry 2000 0.5 <1 ?
Triangulation 2000 0.05 10 100,000
Contact Probe 300 0.05 260,000 (3 days) 2,000
Morié (commercial) 200 0.2 1
Morié (theoretical) n/a 0.001 n/a n/a
Table 1 : A comparison of 3D scanning methods
Proc. SPIE Vol. 4298 107
Figure 4 : Experimental laser stripe triangulation cuneiform scans
5.1. File sizes
We have obtained some preliminary scans using laser stripe triangulation at a resolution of 0.1mm; however, at this
resolution the resulting 3D mesh model is over 100MBytes. It is not practical to transfer datasets of this size over the
Internet, and is very wasteful of bandwidth considering that some scholars only require a low-resolution model. In order to
satisfy the needs of various users, scalable delivery is desirable. A typical scenario would be this. The user first obtains a
model of the entire tablet at a low resolution. The actual resolution will depend on the capabilities of the user’s viewing
platform and on the available communication bandwidth. For example, if the user were viewing the tablet on a hand-held
computer with a slow modem, a very low-resolution model would be transferred in a few seconds. If the user was using a
workstation with a large high-resolution screen, connected to the Internet with a high-speed network connection, then a
much more detailed model could be transferred. If the user chooses to zoom in on a specific area of the model, additional
information could then be transferred to increase the resolution of the area of interest. The exact detail of the zoomed-in
model would depend on the user’s computer capability. The important difference between this technique and standard
delivery mechanisms is that the user would not need to download an entire high-resolution model in order to zoom in on a
There are emerging standards that may provide assistance with scalable delivery. The MPEG-4 standard11 allows for
scalable rendering of 2D objects, but this is mainly aimed at multicast or broadcast data. It does this by delivering the entire
dataset across the communication channel; the users' rendering device can then discard data if a low-resolution model is
required. This technique is not suitable for delivery of specifically tuned data to individual users over narrow-bandwidth
channels, which is what is required for our needs. The Synthetic/Natural Hybrid Coding group12 is currently investigating
mechanisms that achieve interoperability and scalability of mixed media types suitable for different storage media and
communication bandwidths. The group is also working on server-based view-dependent terrain or object rendering and
view-scalable audio-video in 2D/3D environments. To date, the work on mesh coding has focused on 2D models, but there
are plans to extend the work to 3D mesh coding in the future.
5.2. Object tagging
It is very important that the 3D datasets are traceable throughout the entire capture, storage, delivery and manipulation
process. There are many important pieces of information associated with the electronic representation of each tablet, for
example the catalogue number of the cuneiform tablet and information on the capture process. Other information such as
size, excavation details, storage location, ownership, and translations etc. may need to be included in the data tag. As the
user manipulates the data and zooms in, information on the physical size and location of the subset of the dataset should be
available. Since every user may have a different dataset, depending on their needs, this tagging must be embedded in the
data so that it survives further manipulation.
Proc. SPIE Vol. 4298
It is also important that the data tags are available independently of the data itself and the tags should be accessible in an
electronically searchable format. A typical use of this may be a user wanting to find all tablets excavated from a certain site,
and they should be able to search the data tags and obtain a list of relevant tablets. The user could then select tablets of
interest for further examination.
5.3. Issues of copyright
A further requirement may be access control to certain tablets for copyright reasons. Some tablets may only be visible to
authorized users, and this authorization should somehow be carried with the dataset representing the tablet. Another use of
access control would be to limit the available resolution depending on user privileges. Public access could be given to low -
resolution data, and access to the high-resolution models limited to forensic experts. To prevent copyright infringement, it is
important that this copyright information is closely coupled to the data itself. Techniques such as watermarking could be
used, although there are no standards for 3D watermarking data at present.
The MPEG-7 standard11 may offer a potential solution to this problem, but the standard is not yet complete and it is unclear
if it offers the flexibility for our specific needs.
5.4. The virtual reality modeling language (VRML)
VRML13 is a modeling language that can provide an environment for manipulation of 3D datasets. Standard VRML
viewers work by downloading a complete dataset and rendering this on screen depending on the user’s chosen viewpoint.
When used as a mechanism to view 3D models such as cuneiform tablets, the user can select which part of the tablet to look
at (two-dimensional x and y co-ordinates) and can zoom in and out (z co-ordinate). Also, the user needs to be able to rotate
the tablet in all three axes to be able to see all of the faces of the tablet. This gives six degrees of freedom; but most
computers have only two-dimensional input devices (mouse, trackball or joystick). To provide the requisite six degrees of
freedom, VRML viewers have different modes such as ‘pan’, ‘turn’ and ‘roll’ in order to translate movement of the input
device into the required movement of the image. For example, the Cortona VRML viewer14 has three major modes (‘walk’,
‘fly’ and ‘study’) and four sub-modes. Figure 5 shows the Cortona viewer displaying a sample 3D laser scanned cuneiform
Standard VRML interfaces can be confusing,
particularly to novice users, if, for example, left-right
movement of the input device produces rotation of the
tablet, when the user was expecting a lateral
movement of the viewing window. An additional
complexity is added when illumination is taken into
account; for correct viewing of the cuneiform tablets
the user must be able to alter the position of one or
two light sources to create the shadows to provide the
necessary depth information. This gives two or four
extra parameters that need to be manipulated. An
ideal solution would be a form of data glove that
measures the position of the user’s hand (in all six
degrees of freedom) and renders the image to
correspond with the position of the hand. The other
hand could be used to manipulate the position of the
virtual light source using another data glove. If more
than one light source is required, a more flexible
human input device would be required.
Figure 5 : Cortona VRML viewer
Proc. SPIE Vol. 4298 109
5.5. Data compression of 3D sources
In order to store and deliver the large datasets we require, efficient and scaleable compression techniques are essential.
Scalable delivery of 2D data is incorporated in several image formats (e.g. JPEG 2000) and some of these use wavelet
algorithms to generate good compression ratios. G. Taubin15 has developed techniques for 3D geometry compression and
progressive transmission of polygonal models highly relevant to these objectives.
6. CONCLUSIONS AND FURTHER WORK
Our experiments have indicated that sufficiently high-resolution scans are not easily realisable, and that in any case the
resulting file sizes would prohibit remote access and real-time manipulation. Experimental scans were performed using
laser stripe triangulation at resolutions of 10 lines per millimeter producing files in excess of 100MBytes. The current cost
and complexity of the appropriate scanning processes makes remote capture, formatting, storage and communication even
more challenging. The next phase of work will include the application of scaleable 3D compression methods and further
investigation into robust and detailed data tagging systems.
Our sincere thanks to Ron Spencer of Mechanical and Manufacturing Engineering, The University of Birmingham,
Professor Paul Ward of Kodak, UK and also to 3DScanners of Coventry. Also very special thanks to Graham Norrie for his
very valuable photographic advise and to our many project students, in particular, in the School of Electronic and Electrical
Engineering ; M. H. Yu (Ben), Matt Atkins, Peter Cooper. Last, but not least, our sincere thanks to Dr J. Curtis, Keeper of
Western Asiatic Antiquities and his staff at the British Museum for their support and co-operation.
1. C. B. F. Walker, Cuneiform (Reading the Past, Vol 3), Univ California Press; 1989, ISBN: 0520061152
2. J. Nasmyth, James Nasmyth Engineer - An Autobiography, John Murray Pub., London, 1885
3. T.N. Arvanitis, T.R. Davis, N. Hayward, A. Livingstone, J. Pinilla-Dutoit, S.I. Woolley, The Work of the Cuneiform
Digital Forensic Project, VAST: Virtual Archaeology between Scientific Research and Territorial Marketing, Arezzo
(Italy), 24-25 November 2000
4. A. Livingstone, S.I. Woolley, T.R. Davis, T.N. Arvanitis, XML and Digital Imaging Considerations for an Interactive
Cuneiform Sign Database, Electronic Publication of Ancient Near Eastern Texts (Conference), The Oriental Institute of
the University of Chicago, October 8-9 1999
5. N. Demoli, U. Dahms, H. Gruber, G. Wernicke, Optical pattern recognition in the analysis of ancient Babylonian
cuneiform inscriptions. Holography and Correlation Optics II, SPIE-Proceedings 2647, pp. 138-144
6. N. Demoli, U. Dahms, H. Gruber, G. Wernicke, Cuneiform recognition experiments: Coherent optical methods and
results. In: Optical techniques in the humanities, D. Dirksen, G. V. Bally (eds) Springer 1997, pp. 171-174
7. The Digital Michelangelo Project, Stanford University, USA http://graphics.stanford.edu/projects/mich/
8. 3D-Matic Research Laboratory, The University of Glasgow, United Kingdom http://www.faraday.gla.ac.uk
9. 3D Scanners Ltd, http://www.3dscanners.co.uk
10. X. Xie, J.T. Atkinson, M.J. Lalor, D.R. Burton, Three-Map Absolute Moiré Contouring, Applied Optics, Vol. 35, No.
35, 10 December 1996
11. The Moving Picture Expert Group, http://www.cselt.it/mpeg/
12. The Synthetic Natural Hybrid Coding Group, http://www.sait.samsung.co.kr/cgi-bin/snhc/src/index.html
13. The Web3D Consortium, http://www.vrml.org/
14. The Cortona VRML client, Parallelgraphics, http://www.parallelgraphics.com
15. G. Taubin, Visual and Geometric Computing Group, IBM T.J.Watson Research Center,
Proc. SPIE Vol. 4298