Keywords

1 Introduction

3D technology is widely used in different fields of archaeological research (Lieberwirth and Herzog 2016), as means for the documentation of archaeological excavations or historical buildings and even more for recording entire landscapes by remote sensing. One field, the documentation of artefacts in concert with computer-aided data analysis, seems the least pronounced one in this popularisation of 3D archaeology. We will focus in this paper on this specific field, more precisely, we will describe 3D technology and search and exploration methods applied in ancient pottery research.

Ancient pottery, so-called “vases” in archaeological terms (Boardman 2001), belong to one of the largest categories of physical remains of ancient cultures, due to the relative durability of its material. Since the eighteenth century, special attention has been given to this category, especially to Greek painted pottery (Flashar 2000), not only as objects of archaeological research but also as a collector’s item (Nørskov 2002). In archaeology, the artistic analysis of the vase painting was often the focal point until late in the twentieth century by neglecting the three-dimensionality of the object. Today, the research questions focus more on the relation between shape and figured depiction, on the iconographic changes during times, on the content and context of the vases, and many more. Overall, the research on Greek vases became a part of the emancipating Material Culture Studies which focuses on the relations between human and object (Langner 2020).

Regardless of the kind of the research questions, an intensive investigation of the vase should be the starting point for each further discussion. This should be undertaken at the best by autopsy, but to explore each single vase relevant for the respective study at first hand is almost impossible due to the world-wide distribution of this material. Hence, an appropriate publication should deal comprehensively with the vase which includes a full range of measurements, detailed photos, unwrapping where necessary, and an extensive verbal description. Scientific analyses could be added to answer some specific questions, e.g., to get information of the provenance by analysing the used potter’s clay, to identify organic markers relating to potential content, to characterise older restorations or even to date the ceramic material.

In the archaeological domain, pottery is usually published in printed media which force the three-dimensionality of the object into a two-dimensional figure. The standard reference for Greek pottery is the Corpus Vasorum Antiquorum (CVA), an international research project for the documentation and publication of ancient ceramic from museums, universities and other collections. Since the first volume of the CVA in 1922 more than 400 fascicles have appeared, with more than 100,000 vases. Next to the CVA stands the Beazley Archive Pottery Database (BAPD), a freely available online database of mostly Greek vases (c. 120,000) which allows simple searches and filtering (Mannack et al. 2024). The CVA as well as the BADP are still growing.

This historically developed practice of pottery publications is well-established, but cannot do justice to all the research questions on pottery. With the advent of new digital technologies, contactless 3D measurements using optical scanners and X-ray imaging procedures as Computed tomography (CT) were introduced to establish 3D models of vases with the basic aim to create a more objective and complete documentation. On a large scale, 3D technology was applied for the first time for the CVA Vienna Kunsthistorisches Museum 5 (Trinkl 2011; laser scanner) and for the CVA Amsterdam Allard Pierson Museum 4 (Van de Put 2006, medical CT). Despite constraints at that time of early 3D technology (e.g., the low resolution of acquired texture data towards conventional photography), these innovative approaches have played a seminal role in this field of digitisation of Greek pottery.

Only in the last decade 3D technologies are capable of creating 3D models of pottery objects with appropriate accuracy in geometry and resolution in texture, thus being of equal value to traditional pottery documentation, e.g., by means of photography or drawings. This advance has paved the way to exhaust more comprehensively the potential of 3D models, not only for documentation purposes but also for 3D data analysis, and to develop new methods for searching, comparing, and visually exploring 3D cultural heritage (CH) objects (see the overview in Karl et al. 2022). However, for comparative studies high-resolution 3D models of Greek vases are still rarely available. Therefore, a general aim in this digitisation process of CH objects is to make this data, including all necessary metadata, photos and 3D data, freely available, as it was done by the Online Database for research on the development of pottery shapes and capacities (ODEEG; Lang-Auinger et al. 2021). Additionally, due to the previous publication work on pottery with an extensive quantity of data (mostly photos and drawings), novel ways are needed for a joint exploration of these different modalities.

2 Methods of 3D Data Acquisition of Small-Scale Objects – An Overview

The starting point for any kind of digital analysis is the digitisation of the object. Whatever method used, the physical object should be cleaned thoroughly at the beginning. If possible, modern additions, like complemented parts or overpainting, should be removed and further conservation treatment (Kästner and Saunders 2016) should be limited to a minimum.

3D data acquisition methods are based either on direct measurements (e.g., via a laser beam or by triangulation using structured light), on photogrammetry or on X-ray volume reconstruction technology. For the documentation of Greek vases, laser scanning (Hess 2017) was used in the beginning. With advancing technology Structured Light Scanning (SLS) is currently widely used in pottery studies (Rieke-Zapp and Royo 2017). Both techniques are optical methods and ensure the acquisition of precise and accurate geometric data of the ceramic surface. Within the last decades, Computed Tomography (CT) has been developed to be a notable imaging method in the field of non-destructive testing (NDT), enabling dimensional measurements and material characterisation (Carmignato et al. 2018). All these methods result in a well-built 3D model, but lack an appropriate recording of the mostly painted vessel’s surface (the texture) aligned to the needs in archaeological research. For acquiring high-resolution photo-realistic surface models which are especially needed for the vase painting multi-image 3D reconstruction like Structure from Motion (SfM) and Dense Multi-View 3D Reconstruction (DMVR) (Koutsoudis et al. 2013; Hess and Green 2017) is currently the most effective solution. The combination of acquisition methods using the strengths in each case has been proven to be leading to the best results (cf. Sect. 7.3.4). A general overview for all kinds of scanning methods, including also the potential and limitations is given by Dey (2018).

All of these techniques share the same overall concept of contactless measuring (no physical touching of the surface) which guarantees an optimised data output by minimising the risks of damage or even (partly) loss of the archaeological substance.

3 Applications in Pottery Research – Case Studies

In the following, we will present selected case studies in the field of computer-aided Greek pottery research conducted by the authors and collaborators associated with the CVA community. They are based on diversely acquired digital data and develop novel approaches for further academic discussions.

3.1 Unwrappings of Painted Curved Surfaces

A fundamental task of high significance in research on ancient vase painting is the unwrapping of the painted vase surfaces (Walter 2008). These unwrappings show the depictions with minimal photographic distortions or sectioning by separate photos, enabling archaeologists to analyse and interpret the image as a whole in terms of style, dating and iconography. They are typically created manually using tracing paper, which is time-consuming, error-prone, and often not even allowed due to the required contact with the fragile surfaces. Another method, peripheral or rollout photography, is contactless but can only be applied reasonably for cylindrical painted surfaces (Villard 1965; Felicísimo 2011).

Today, various 3D mesh processing and visualisation tools, like the GigaMesh Software Framework (Mara et al. 2021) or CloudCompare (Girardeau-Montaut et al. 2021), allow to perform such unwrappings directly on a virtual 3D model of a vessel (Rieck et al. 2013; Karl et al. 2019). They utilise proxy geometries that exhibit a simple surface of revolution (cylinder, cone, sphere) that best approximates the vessel shape. This proxy is computationally fit to the 3D mesh, which is then unwrapped according to the unrolling of the proxy around its axis of revolution. The resulting rollout can then be projected to 2D, for instance, along an overall optimally orthogonal angle. This results in a “flat” representation displaying the entirety of the vessel surface, but can show considerable distortions in stronger curved surface parts (Fig. 7.1).

Fig. 7.1
A photograph and 2 models of a painted vase. a. A photo of a vase painted with a figure that has a human head and a snake's body, surrounded by flower-like patterns. b. A 2-D unrolled model with a rectangular shape. c. A 2-D unrolled model in the shape of a flattened cone.

Computer-aided rollouts of the Corinthian alabastron University Graz G28: (a) photo; (b) cylindrical; (c) conical rollout. (© P. Bayer, S. Karl, J. Kraschitzer, University of Graz)

Full size image

A major issue with these kinds of unwrappings is that unless dealing with purely developable surfaces, the projection to 2D will necessarily introduce different types of surface distortions. Conformal methods strive for preserving angles, that is, avoiding shearing of surface motifs, but can introduce strong undesirable distortions of distances and scale (cf. mapping of the earth: Snyder 1993). In contrast, distance preserving methods introduce strong angular distortions that can render the result useless as well. Especially for pottery objects that exhibit highly curved, bulky shapes, the effects of this mapping problem can become practically problematic in the attempt of creating an all-encompassing depiction of the surface paintings that is true to scale in all relevant details.

To address this problem, more elaborate mapping techniques can be employed that minimise a defined distortion error measure (Floater and Hormann 2005; Sheffer et al. 2006), e.g., using a numeric optimisation on an initial mapping. Starting from a naive unrolled surface with potentially strong distortions (Fig. 7.2b), the Elastic Flattening (EF) approach (Preiner et al. 2018) computes a physics-inspired relaxation of the stresses induced by these distortions on the edges of the 3D mesh. In this process, mesh vertices are iteratively relocated to minimize the deviation of the length of each edge in the planar map from its original length in the 3D mesh. This way, the introduced distortion error is distributed evenly over the surface. As seen in Fig. 7.2c, the resulting depiction is able to significantly reduce both proportional and angular distortions compared to the naive initial rollout. It has also been shown that the EF results widely agree with the layout resulting from manual unwrappings of comparable vases (Fig. 7.3).

Fig. 7.2
A photograph and 2 unrolled models of a painted vase. a. A photo of a vase with 2 handles with a painting of 2 persons. One person sits on a chair with wheels and wings, while another holds a jug. b and c. The distorted face of the seated person has a box. The standing person has angular distortion.

Attic red-figure hydria, University Graz G 30; (a) photo; (b) spherical rollout exhibiting proportional (yellow) and angular distortions (white); (c) Elastic Flattening. (© Preiner et al. 2018, The Eurographics Association)

Full size image
Fig. 7.3
A flattened 2-D model of a painted vase and a drawing. The painting is of a figure that has a human head and a snake's body, surrounded by flower-like patterns. The model and the drawing have similar curved shapes.

(a) Elastic flattening of the Corinthian alabastron University Graz G 28 in comparison to (b) a hand-drawn unwrapping of the alabastron Brussels R 224 with comparable motiv from the same vase painter (Lenormant and de Witte 1858, pl. 31). (© R. Preiner, TU Graz)

Full size image

This work on optimal digital unwrappings of Greek pottery raises the potential for further research. In contrast to naive unwrappings that produce divisive cuts through different motif parts (e.g., neck of the bird in Fig. 7.3a), future improvements will involve finding optimised layouts that preserve the integrity of the motifs, which is of primary importance for the archaeological interpretation.

3.2 Shape Comparison

The spatial expansion, the geometry, is among the most significant features of a vase. Shape was always used as classification criteria for establishing typologies. Hence, digital geometric analysis started early, cf. the overview by Pintus et al. (2016), mainly focusing on sculpture (Lu et al. 2013; Frischer 2014) and terracotta (de Beenhouwer 2008).

Whereas the vast majority of the Attic pottery is thrown on the potter’s wheel, there is a production of mould-made Attic vessels from the late sixth and fifth century BC, preferably in the shape of a human head, so-called head vases. Replicas of the same mould can be identified by using 3D models and computer aided matching (Trinkl and Rieke-Zapp 2018). The difference between similar head vases can be quantified. It enables the detection of a series that is taken from a single mould (Trinkl et al. 2018). Furthermore, by comparing similar head vases with different heights, at least three interdependent series are evident (Fig. 7.4). This can be explained by the manufacturing process of re-molding, which results in copies of progressively smaller height due to the shrinking of the clay during the drying and burning process. The use of digital 3D models also enables the evaluation of fragmented objects, which is hardly possible by an analysis using conventional measurements.

Fig. 7.4
27 models of vases with the shapes of similar human heads at the center and handles on top. Row 1 has 7 models, and rows 2 and 3 have 10 models each. The vases are indicated by shades for 9 museums in Berlin, Bologna, Budapest, Ferrara, Florence, Nicosia, Tuebingen, and Vienna.

Three interdependent series of head vases stored in nine different collections. (© P. Bayer, E. Trinkl, University of Graz)

Full size image

3.3 Filling Volume Calculation

The shape of a vase and its filling volume are closely related. The determination of the filling volume is essential to detect standardisation in the potters’ production and to recognise ancient units of capacity which varied according to location and epoch (Büsing 1982).

If a vase is unbroken and well preserved, the capacity can be measured indirectly by filling the vase with dry granular substances, like rice or sand, and then measuring the capacity of these decanted substances. However, as in practice most vases are too fragile, a contactless measurement has to be performed. For so-called “open vessels”, i.e., vases whose inner surface is visible and can therefore be measured, the easiest way is to rotate the measured inner profile and calculate the volume of the resulting body of rotation (Moreno et al. 2018). A web application developed at the University of Brussels provides this computation for domain users (Engels et al. 2013; Tsingarida et al. 2021). Whereas this calculation is based on assuming the volume to be a body of rotation, only 3D scanning can completely capture the inner surface of open vessels and allow an accurate calculation of the filling volume.

With 3D models it is also possible to estimate the inner surface of so-called “closed vessels” (e.g., Fig. 7.5a), i.e., vases of which the inner surface cannot be measured, e.g., because of a narrow mouth. Based on prior knowledge of the wall thickness, an offset of the outer surface towards the interior can be determined to estimate the filling volume (Mara and Portl 2013).

Fig. 7.5
Four 3-D cross-sectional models of a vase with a thin neck and a broken handle. a. The vase interior is in a single shade. b. The vase has 2 vertical sections in different shades. c. The left section has a painted exterior, with the interior on the right. d. Horizontal striations inside the vase.

Corinthian alabastron, University Graz G 28: (a) Isosurface volume rendering of CT data (transparent modus); (b) CT surface with incisions, one half enhanced by using Multi Scale Integral Invariant filtering; (c) textured CT surface with sectioning; (d) volumetric “phantom” body of the capacity (1,493 ml). (© S. Karl, University of Graz)

Full size image

A more complex method for estimating the filling volume of closed vessels is again based on the scanned outer surface, but utilises the mass of the vessel and the bulk density of the ceramic material to calculate the ceramic volume and thus the wall thickness (Spelitz et al. 2020). The material density can be determined from a pottery fragment with the same material properties, so-called “fabric”. Unfortunately, the majority of the vases in museums are restored and completed with other materials, which affects their mass. In general, the determination of bulk densities as characteristic properties for specific fabrics (e.g., Attic or Corinthian) is still at the beginning and requires more large-scale test series (Karl et al. 2013).

The most precise method of receiving the filling volume of closed vessels is to use the 3D data acquired by Computed Tomography (Fig. 7.5d), which, however, requires expensive stationary hardware and is thus less accessible to most domain users.

3.4 Identification of Manufacturing Techniques

Besides shape (geometry) and decoration (texture), manufacturing techniques provide other attributes to classify and interpret pottery; hereby focusing on the choices and changes in technical practices (Rice 2015). In wheel-thrown pottery, which most Greek pottery belongs to, traces of primary manufacturing techniques such as potter’s finger striation marks or the location of joints of separated formed parts are mostly preserved in the interior of closed vessels or on subordinate parts. On the exterior, these traces are usually eliminated by secondary smoothing and burnishing, finally by painting.

For this field of pottery analysis, X-ray imaging methods, recently CT were used (Van de Put 1996; Kozatsas et al. 2018). A particular strength of this method is that visualisation and analysis can be performed on the whole vase (Karl et al. 2013, 2014). CT provides an accurate and complete 3D documentation of an object encompassing all internal structures (Fig. 7.5a); even fine details such as the incisions of the black-figure style can be displayed due to the high resolution (Fig. 7.5b). The CT model can be additionally combined with texture information, e.g., acquired from an SfM model (Fig. 7.5c). Based on the recording of the object’s interior surface, the vessel’s capacity can be calculated with high accuracy (Fig. 7.5d).

While the use of the potter’s wheel can be clearly identified by the elongation of voids and other inclusions in a spiral pattern (Fig. 7.5a), separately attached vessel parts are mostly recognised by the change in the structure within the ceramic body. Furthermore, the CT data allows to reveal traces of used pottery tools, ancient repairs during the manufacturing process (Karl et al. 2018) or modern interventions and additions.

A unique point of CT compared to all other methods is the fact that it is able to “look” into the material without cutting it (Fig. 7.6). Depending on the accuracy of the CT scan, it enables a detection and morphological analysis of the air pores (voids) and inclusions within the ceramic matrix (e.g., according to amount, size, shape). Matrix is commonly termed the fine micaceous basic substance of the burnt clay, while inclusions are so-called non-plastic components, mostly originating from tempering the potter’s clay. The fact that these inclusions become visible at all is due to the complex assemblage of the ceramic material, which consists of mineral particles of different specific gravity, e.g., clay minerals, quartz, feldspars or iron oxides. A quantification of the clay fabric properties enabled by this non-destructive method allows for a material characterisation, which is an important methodology in pottery research (Gassner 2003), not only for questions of manufacturing technology but also for the localisation of the production site or the workshop.

Fig. 7.6
A photograph, a 3-D model, and a C T scan image of a square fragment. a. A photo of a fragment painted with the legs of people and patterns. b. A 3-D model with pores in different shades. c. The fragment is in a light shade with pores in a dark shade.

Fragment of an Attic Late-Geometric krater, University Graz G 517: (a) 3D model; (b) 3D visualisation of porosity (connected voids colored), (c) CT cross-section with voids (black) and different inclusions (middle grey and white). (© S. Karl, K.S. Kazimierski, University of Graz)

Full size image

Even though CT offers a high potential in documentation and identification of manufacturing techniques, it comes with certain drawbacks. First, the sensitive objects must be transported from its storage location to a specific CT lab, which often requires additional efforts and precautions. Moreover, typical CT artefacts like beam-hardening can affect quantitative analyses and CT surface reconstructions (Carmignato et al. 2018; Kazimierski and Karl 2015). Future research in the archaeological domain will have to consider the use of mobile and more flexible X-ray imaging devices for achieving adequate information of the vessel’s interior.

3.5 Shape-Based Retrieval

Apart from individual analysis and pairwise comparison, an essential task in pottery research involves the comparison of multiple objects to a query in relation to different similarity traits, e.g., shape, texture, painting style or metadata. Retrieval methods enable to rank the objects in a (possibly huge) database with regard to a given query, which generally consists of keywords, but can also comprise images, sketches, or 3D shape information (Biasotti et al. 2019; Rostami et al. 2018).

In terms of Greek pottery the objects’ shapes are a fundamental trait for comparison. To date, many shape analysis methods have been proposed for applications in CH object data (Pintus et al. 2016; cf. 3.2). The amount of published vases is huge and accompanied with comprehensive metadata and a high number of images, while 3D models are rarely available. Hence, one has to resort to comparing their shapes based on available images depicting their silhouettes, using appropriate image comparison techniques.

These images are compared using mathematical representations of characteristic features of the silhouette, image color patterns, etc. These so-called “feature descriptors” enable the computation of similarity measures between images. The variety of feature descriptors is vast and they can be divided into engineered features, based on explicitly defined transformations of the input images, and learned features which are relying on machine learning algorithms.

Suitable similarity measures have been obtained e.g., by the engineered Histogram of Oriented Gradients (HOG) (Dalal and Triggs 2005) feature descriptor, which encodes the orientation and magnitude of the color gradients over pixel blocks. An alternative is given by the Shape Contour Descriptor (SCD) (Attalla and Siy 2005) which is solely based on the silhouette of a depicted object.

State-of-the-art methods also allow to search for similar vases given only fragmented or incomplete vases, by sketching the supposed completed silhouette in a graphical user interface (Lengauer et al. 2020). As shown in Fig. 7.7, these methods provide a high success rate even in case of fragmented query objects.

Fig. 7.7
62 photographs of vases with long necks and a handle. The reference image has a complete vase. The query image has a broken vase. Row a has 20 similar vases. Row b has 16, 14, 4, 13, 2, 1, 11, 3, 17, 5, 12, 19, 6, 18, 10, and 7 ranked and shaded. Row c has 15, 3, 14, 17, 7, 28, 19, 16 and 27 shaded.

Shape based retrieval with HOG (b) and SCD (c) descriptors compared to a manual expert ranking (a). Each row shows the ranked top 20 results for a fragmented sample query on a diverse database with 3,340 object depictions. (© S. Lengauer, TU Graz)

Full size image

3.6 Motif-Based Retrieval

Apart from shape, the ornaments and figural depictions, the motifs, on the painted vases are often an important basis for the analysis and exploration of ancient Greek pottery. These motifs are manifold and include single figures as well as multi-figured scenes (Fig. 7.8a), e.g., deities, mythological figures, weddings, sacrifices or warrior departures.

Fig. 7.8
12 photos and 24 processed images of vases. a. 12 photos with multiple views of 10 vases of different shapes, sizes, and paintings. b. The outlines of the vases, backgrounds, and painted figures are indicated in different shades. c. The shaded figures are against dark backgrounds.

Segmentation examples for a set of images of painted vases (a) with EGBIS (b) and morphological segmentation (c). (© Lengauer et al. 2019, The Eurographics Association)

Full size image

From a technical perspective, the challenge of finding vases with similar motifs can be split into two major parts: (1) An image segmentation part for composing a database of motifs and (2) a matching part determining the similarity of all motifs in the database to a provided query (Lengauer et al. 2019). Image segmentation describes the process of assigning the pixels of an image to a finite number of coherent regions. For the task of extracting motifs from a picture, those regions should ideally correspond to the individual motif outlines. We have obtained good results in our work with the Efficient Graph-Based Image Segmentation (EGBIS) algorithm (Felzenszwalb and Huttenlocher 2004; Fig. 7.8b) as well as with segmentations based on morphological transformations (Fig. 7.8c).

In the study of vase painting, it is generally accepted that similar motifs have comparable outlines or contours. A feature descriptor like Shape Context (Belongie et al. 2002) represents an appropriate choice for quantifying the similarity of outlines extracted by segmentation to a given query. As shown in Fig. 7.9, this approach allows to find and discriminate similar motifs.

Fig. 7.9
22 photographs of vases. a. A vase that has a painting of a winged figure with an outstretched arm, and similar paintings outlined and highlighted on 10 other vases of different shapes. b. A vase with a painting of a flying figure with wings, with similar paintings outlined on 10 other vases.

Motif retrieval examples of (a) a standing figure with outstretched arm and (b) a winged flying figure, the Eros, with the sorted top results for these different user-defined queries. (© Lengauer et al. 2019, The Eurographics Association)

Full size image

We find that a successful segmentation for this motif-based approach is often hindered by the degeneration and incompleteness of the vase surface (e.g., in case of erosion) and by the interlinking and overlapping of motifs.

3.7 Multivariate Structuring of Large Object Collections

A central task in archaeology is the classification of objects according to various object properties (Adams and Adams 2008). While individual objects are typically classified via similarities to known objects, large collections of (digitised) objects represent a much more tedious task for classification, which typically starts with organising the objects according to their numerous properties (e.g., date, findspot, shape, etc.) and goes further to building groups with common properties. Important insights are mainly based on analysing the relations between these groups, e.g., temporal clusters that are related to object accumulations in a particular site. However, revealing these relations by manual investigation is a highly complex task.

Appropriately designed computer-aided visual analytics tools can greatly support archaeologists in organising and grouping objects with respect to date, findspot, and shape, and allow to visualise significant relations between groups within these different dimensions. Different properties can be assigned to different spatial dimensions in an interactive three-dimensional system (Windhager et al. 2020). Network visualisations are an established base technique to illustrate object relations (Van der Maaten et al. 2007; Bogacz et al. 2018) and can also be combined with additional visual metaphors for particular properties, e.g., displaying time as a temporal landscape (Preiner et al. 2020).

An integrated linked view system such as the Linked Views Visual Exploration System (LVVES) depicted in Fig. 7.10, allows the coherent exploration of findspot, date and shape information. This is facilitated through a separate viewer for each of the mentioned properties (Fig. 7.10), consisting of a map for the findspot, a timeline for the date and a network visualisation for the shape information. While the structuring of objects within each view allows for an exploration within a single dimension, an additional intra-view linking mechanism allows to highlight objects in all other views, revealing relations between groups across dimensions (red connections in Fig. 7.10). This approach is not limited to these three properties but can be extended to display additional characteristics like painting style, fabric, and more.

Fig. 7.10
5 screenshots of pop-up windows. The windows titled G M V have maps of countries, with images of vases marked on different locations. The windows titled S S V have images of vases interconnected by lines. The windows titled T V have horizontal floating bar graphs with shaded rectangles indicated.

LVVES, visualising a selection of objects structured by findspot (GMV), shape similarity (SSV), and date (TV). Intra-view connections (blue rectangle) are revealed through linking and highlighting mechanisms. (© S. Lengauer, TU Graz)

Full size image

4 Discussion and Outlook

Once generated, the benefit of a 3D model is wide-ranging. The digital model may be used, re-used and modified as many times as wanted, without touching the original object again. Using non-tactile acquisition techniques, the protection of fragile objects or objects of poor preservation is provided in the best possible way. A digital documentation can enrich the conventional measuring and description; extend visual capabilities (cf. Sect. 7.3.1), supports quantified surface comparison (cf. Sect. 7.3.2) and enables calculation of capacities (cf. Sect. 7.3.3). Depending on the used methods and tools it even offers insight into the material properties (cf. Sect. 7.3.4).

In particular, the presented case studies demonstrate that vases stored in diverse locations can be compared easily without being moved (cf. Sect. 7.3.2); moreover, partly preserved vases can be included in the evaluation. A digital environment simplifies comparisons of single features of the vase, like shape or motif (cf. Sects. 7.3.5 and 7.3.6), and the linking of features like chronology, findspot and shape (cf. Sect. 7.3.7). By this means new relations can be revealed and already known relations can be visualised.

Additionally, to the above presented analyses of object’s properties like geometry and texture, further scientific approaches associated with 3D data can reveal object’s properties that cannot be detected by traditional archaeological practice. A very valuable method is the combination with non-visible light (UV, IR) for the detection of conservation details and recent manipulation (Kästner and Saunders 2016; Nocerino et al. 2018).

Conveying the manifold information and complex meaning of Greek vases to non-archaeologists can be difficult. Hence, a 3D model may be applied in the dissemination of expert knowledge to make our common CH more familiar to a growing audience (Quattrini et al. 2020). For various kinds of dissemination, a replica based on a 3D model can be useful, e.g., in exhibitions and in classrooms on various levels of education (Breuckmann et al. 2013).

4.1 Challenges

Despite the various prospects of digitisation for the analysis and documentation of vases showcased above, their usage and utilisation for practical archaeological tasks faces several challenges.

The acquisition of the data oftentimes requires special hardware and associated skills for their operation. Moreover, certain digitisation equipment can be rather expensive, others are rarely available and often not mobile. These factors have to be considered when discussing the documentation costs. Furthermore, a future utilisation of the data in new or upcoming archaeological research questions requires defining the detail, quality and nature of the data already at the time of acquisition, which is difficult to anticipate. Approaches for mass digitisation which can be configured for different acquisition modalities, may provide a scalable digitisation infrastructure (Santos et al. 2014).

The preservation of the data itself often comes with considerable long-term storage costs, and has to handle the choice of suitable and accessible data formats and resolutions. Moreover, it is essential to augment the data with suitable meta information that document the nature and parameters of the acquisition process, to ensure their traceability and interpretability.

Once stored, the retrieval of the data, i.e., its computer-aided search and analysis, requires a scalable and well-structured data pool. 3D data, especially from Greek vases, is rarely available in a structured format, and often lacks a complete set of associated metadata. This, however, is an essential condition for a research-based approach. For the specific field of Greek pottery there is still a lot of work to do on aligning the domain ontology (Gruber and Smith 2015). In 2017, a repositorium established at the Institute for the Study of Ancient Culture at the Austrian Academy of Sciences made a start by creating the first publicly accessible database for ancient vases (ODEEG; Lang-Auinger et al. 2021).

4.2 Outlook

A main future objective is to enlarge the 3D data volume of digitised Greek vases. Only then the presented analyses and computer-aided exploration can display their full impact in archaeological research. Of course, any development of new digital methods has to consider the integration of the huge amount of existing documentation in previous archaeological publications going back to the nineteenth century, mostly only available in images and text. Novel applications may include cross-modal exploration considering diverse modalities like 3D data, photos, drawings, sketches, metadata at the same time. Thereby, computer-aided methods can help additionally to improve existing documentation in 2D or 3D by measuring data quality (e.g., according to shapes, images or text) and by revealing research/documentation needs.

An interesting outlook is also the introduction of advanced machine learning (ML) methods to the field of Greek pottery studies (cf. Langner et al. 2021). The work described above currently rely on so-called engineered features, which use techniques of traditional image and shape descriptions and segmentations. These approaches are well-understood and in our experience robust in many cases. However, engineered features may be outperformed by learned features, e.g., for retrieval or shape completion tasks (Schreck 2017). In our experience, a challenge is how to extract learned features, given that such approaches require training data and choice of learning architecture and parameters. Training data may be sparse in the domain. More research to this end, e.g., in applying existing ML methods trained for generic images to the archaeology domain using so-called transfer learning, is needed.

5 Conclusion

This paper focuses on the research needs in studying CH objects which also includes Greek pottery (vases), a main working field in classical archaeology. A combination of traditional and computer-aided methods is most suitable for a comprehensive exploration of these objects. The traditional methods like hand drawings and sketches, verbal descriptions and the study of publications can be supported by digital methods in many ways; (1) the documentation of a single vase is enriched by digitisation, e.g., specifically by the use of 3D models; (2) the search for comparable material in a wide range of publications is improved by segmentation and retrieval techniques; and finally, (3) visualisation technologies support effective exploration of object repositories and finding correspondences, and enhance the demonstration of research results in publications.

With the above presented case studies we have shown that digitised object data can be a fundamental enhancement for archaeological research. Some approaches are still at the beginning of their development and need further development and more testing. Above all, the targeted digitisation is a basic requirement to advance archaeological research in the field of Greek vases.