A Semi-Automatic Procedure for the Recognition and Classification of Pieces of Archaeological Artefacts
L. De Napoli, M.L. Luchi, M. Muzzupappa, S. Rizzuti
 Dipartimento di Meccanica, Università della Calabria
Abstract
This work aim to provide some tools to help the archaeologist file fragments of artefacts found in archaeological sites. The data from each fragment are captured with a 3D digitizing system and processed. Some significant geometric parameters are estimated to locate the fragment in an approximate  position on the related virtual vessel stored earlier. Then the number of items found in the site is estimated.
1. INTRODUCTION
As any archaeologist knows the archaeological record is too fragmentary to give us a direct vision of settlements, buildings and objects. Usually the context is lost and it is only thanks to the skills of specialists that it is possible to match together the fragments of the past. Each generation of archaeologists must take care of the recovery, documentation and of analysis of different sets, ranging from geographic maps to the objects themselves. In the Department of Mechanics of the University of Calabria, and in collaboration with the Department of Archaeology and Art History, a project has been started to develop a computer aided tool to simplify and speed up the activity of archaeologists who are skilled in filing ceramic finds. These are a real guide-fossil to process archaeological maps concerning anthropization of a well defined territorial context. In fact, the ceramics, once manufactured, are not subject to destruction and thus they stay permanently on a site, in general as fragments, to indicate how stable the settlement of that particular area was. Traditionally the filing of the artefact is carried out following these five steps. The first collected datum is of geotopographical type and refers to the zone where the remains are located; the others are the thickness of the fragments, the colour of the clay, the type of mixtures and finally the morphology of the pieces, that allows its identification by means of  proper typologies [1]. Afterwards fragments are selected that seem to belong to the same item and their shape is determined by means of the detection of curvatures using predefined characteristic templates and similarity to known archetypes [2]. This is of particular importance when the finds are associated with mass manufacturing and the number of the fragments is high. In such case the work of filing, which is entirely manual, is very repetitive. In this case it is not so important to reconstruct the items from the fragments. Instead the main goal of the archaeologist is to determine the total number of complete items that have been found in a site.
 
The purpose of this work is to provide the archaeologist with a tool in order to automate the counting of archeological finds. The work is related to vase artefacts coming from a very interesting site located in the  province of Cosenza in Northern Calabria. The place called Cozzo la Torre, in the zone of Torano Castello is very interesting from an archaeological point of view, because in this site, settlements can be located from the twelth to the second century BC. The first step of the method consists of the acquisition of point clouds for the extrados of the fragments, by means of a mechanical 3D digitizing system. A procedure for the reconstruction of the artefact surface, by means of the reverse engineering methodology, stores the geometrical data of each piece. In order to facilitate the work of the archaeologist a 3D scanning technique has been adopted to obtain the surface in a sufficiently accurate way. The reconstructed surface of each fragment is compared with the surface of a series of virtual vessels, formerly modelled, until its position is well determined on it. The position of each fragment and its dimensions are thus stored in a suitable format. This operation of matching is repeated for every piece. At the end of this phase a procedure is able to estimate the number of each type of vessel that is present in the archaeological site.
2. THE ARCHAEOLOGICAL SITE
Cozzo La Torre at Torano Castello in the province of Cosenza [3,4] is a site located on a hill, along the west slope of the Crati valley (
 fig 1
). It was a fortified settlement,  populated in the fourth and third century BC by the
 Brettioi
or
 Brettii
or
 Bruttii
), an indigenous calabrian people who for many centuries had  been influenced by Hellenic culture. The data are collected from fragments of three types of Apulian-made artefacts [5,6]:
1 kraters
, open shapes, typically with a diameter of 350 mm at the rim and 350 mm high (
 fig 2
);
 2 skyphoi
, open shapes, typically with a diameter of 130 mm at the rim and 130 mm high (
 fig. 3
);
3 lekythoi
, closed shapes, typically with a diameter of 50 mm at the rim and 230 mm high (
 fig. 4
).
 Fig. 1
 
 
 Fig. 2 - krater  Fig. 3 - skyphos  Fig. 4 - lekythos
 
 Fig. 5
The manufacture of all these types of pottery is very elaborate. In fact, most of these were decorative objects to show status. In particular, the first two (
kraters
and
 skyphoi
) were exhibited in important meetings such as  banquets or symposia connected with wine drinking, and for that reason they were a sign of a high social position; these artefacts were so precious that they were enclosed in the funerary kit to display wealth. The third item, the
lekythos
, was instead a receptacle for perfumes, and had always been a fundamental funerary item. The kraters under consideration are red-figured often decorated with various scenes, most of a mythological type, on the two opposite sides and separated by ornamental subjects such as  palms, gyrals and acanthus leaves. The
lekythoi
 are also red-figured but they have a dark whitewashing inside. The
 skyphoi
 are also like these.
3. LITERATURE REVIEW
Geometric modelling is a computer aided technology extensively used in diverse scientific and engineering application domains. Geometric models can be created with various commercial CAD/CAM systems, both feature based and with free-form regions. Often we need to create geometric models of existing objects for which no models exist and that are not easy to define with a typical CAD system. In general, the problem is: given partial shape information of an unknown surface, construct, as far as possible, a complete representation of the surface of an object. This reconstruction process is often referred to as “reverse engineering” [7,8]. There are many different methods for acquiring shape data (point cloud data, in general), as shown in figure 5. There are contact and non-contact methods. In the first type, the surface is touched using mechanical probes at the end of an army (as in this case) or a coordinate measuring machine (CMM).  Non-contact methods can be magne-tic, acoustic or optical (triangulation, ranging, interferometry, structured lighting and image analysis). These are probably the most widely used and most popular with relatively fast acquisition rates. The main purpose of reverse engineering is to convert a discrete data set into a piecewise smooth, continuous model. The discrete data set typically consists of (
 x,y,z 
) coordinate values of measured data points, scattered or regular (depending on the specific set-up). The first step of data processing is the segmentation [9], that allows us to label points  belonging to a particular region, some of which can match surfaces of a particular simple type, like planes, cylinders or others [10,11,12]. Each segmented region is processed to reconstruct a surface. Surface modelling techniques obtain surfaces either by polyhedral [13,14] or curved surface approximation. In this case, surfaces are approximated algebraic methods [15,16], parametric methods (B-spline, Bezier surfaces or NURBS [17,18]) and recently less conventional methods allow a more flexible and suitable handling of surfaces [19,20].
DATA ACQUISITION METHODS NON-CONTACT METHODSTACTILE METHODS OPTICALACOUSTICROBOTIC ARMSMAGNETICCMMsTRIANGULATIONRANGING INTERFEROMETRYSTRUCTURED LIGHTING IMAGE ANALYSIS
 
 Fig. 7  Fig. 6 
 In many cases it is not necessary to define the surfaces of the observed object completely, but it is important to characterize or recognize it. In this case, it can often be sufficient to estimate some geometric characteristics from the point cloud data set. There are many techniques, depending on the specific task. However, often Gauss map is estimated [21], either to identify single points or small localized zones of a surface [22] or to recognize the typology of a part of a surface [23]. The recent theoretical and technological breakthroughs in mathematical surface modeling and data-capturing make it possible to advance 3D knowledge into new realms of cross-disciplinary research, both for a pure processing of the data [24] or for filing and classification of items that can be looked up, also via internet [25]. Particularly interesting is the “Digital Michelangelo Project” [26], in which a cross-disciplinary staff have built a digital record of some of Michelangelo’s statues (David, for example). Many other digital libraries are being implemented for diverse purposes. In particular, at the Arizona State University, there are some projects for archiving archaeological finds in general [27], Native American pottery in particular [28], and for archiving rocks and bones [25].
4. METHODOLOGY
The procedure developed can be classified as a reverse engineering methodology. In order to validate it two sets of points have been considered: one point data set collected from virtual fragments derived from models of vessels created by a parametric CAD system (
 fig. 6 
); the second point data set from the surface of real fragments, collected with a 3D mechanical digitizing system, the Microscribe 3DX
, that operates with an accuracy of
±
0.23 mm (
 fig
). The virtual vessels, created with the parametric CAD system, also constitute the file of templates that must be consulted in order to compare the estimated geometric characteristics of the fragments with the same characteristics present on the vessel. The procedure, that follows the flow chart reported in figure 8, has  been written in
 Mathematica
 4.0. It approximates the point cloud data set with a polynomial surface f(x,y) (
 fig 9
), so the surface can be represented by a single value function with respect to a suitable reference frame associated to the measuring device. In fact the point data set has been collected  positioning the fragment with the Z axis of the coordinate system in such a way that it appears as wide as possible if observed from the Z direction. Besides the X and Y axes must be oriented in order to align to them the sides of the quasi-rectangular  pattern of the point data set (
 fig. 10)
.
of 9