Surface analysis of archaeological artefacts [Draft]

I. Introduction

It has been a while since I posted the last blog entry and this new one is  not exactly an update about my point cloud visualization work.

Recently, following  my work for the Hidden Treasure of Rome project, people asked me to scan and run surface analysis on various potery: from early bronze age production to late roman ceramics.

I thought it was time to share some methodological insights and also some problem which remain to be solved. In this regard, I picked up a very interesting anthropomorphic gallo-roman potery (two fragments) that was submitted to me by a friend.

To give a bit of a context, the fragments were found near a gallo-roman oven (fig. 1) from an artisanal district in La Genetoye monumental area in Autun (France).

Fig. 1: Autun 2014.3 excavations (CAD A. Stock and M. Thivet).

This kind of production is not uknown in the roman empire but is more likely to be found near the eastern border, far from Augustodunum to say the less. Still, we are aware of a few “local” examples and it’s not an unicum.

The scope of the analysis was very broad: gather and highlight as much information as I could from the surface to discuss the method of production, and to some extent, give some arguments to push forward the idea that the two fragments are from the same vessel.

I will focus on the surface analysis and leave the real specialists speak about color, etc. Each step of the analysis were recorded and are available on Youtube:


II. Sofwares and 3D web viewer

Beside the nurbs revolution, for which I will say a few words later, everything was carried out with free softwares listed below the Youtube video. Namely:

III. 3D Scanner

For this 3D scan, I was working with Q. Verriez and we used the Gom Atos Core 80. The same scanner that was used for the Hidden Treasure of Rome project: Blue light structured light scanner with 0.03 mm – 0.05 mm spacing between points at full resolution.

The partial 3D reconstruction of the ceramic was produced after the analysis, through a nurbs revolution in  Autodesk  Maya. I computed a “best fit circle” from a selection of production marks on the surface of the head fragment with GOM Inspect to get an idea of the shape (fig. 2 and 3).

From there, I carved the best profil I could from the head fragment and used it with the said nurbs revolution, the result was not quite as good as I expected  to be honest.

Fig. 2:  Best fit circle feature (CAD D. Vurpillot).

Fig. 3: 3D reconstruction attempt (CAD D. Vurpillot).


IV. Surface comparison from smoothed models

My first idea was to produce a very clean and smoothed out mesh (let us call it SM) from the original scan (OM) and run a simple surface comparison between the two of them.

With this kind of complex surface it’s really hard to figure out the best way to run a relevant surface wise analysis. Initially, I was under the impression that I should focus on the  inside of each fragments and produce a best-fit paraboloid, which could be used for surface comparison afterwards.  Each best-fit paraboloid production algorithm (two are available in GOM Inspect, with subparameters) has its downsides and it turned out that the results were not that good.

That’s why I tried the “raw” SM method, which should be closer to the original surface while highlighting noteworthy local variation (bump or cavity). As shown in the video, my starting point was mesh decimation (keep in mind those are very high poly mesh to begin with) at 10% (25k faces for the ear fragment and around 62k faces for the head). I felt I should go as far as 5% for the head after careful examination. Then, I smoothed the surface with a very large brush in order to get a clean and regular surface.

Again, results for this specific set of artefacts were not as good as what I was expecting. As you can see in fig. 4, we can barely make out the main shaping marks. Some fine tunning could enhance it a bit, but this approach does not seem to be the best one for those artefacts and I was under the impression I should move to defect maps. Still, I would not discard too quickly this methodology as inefficient, I have seen good results with more “suitable” surfaces.

Fig. 4: Surface comparison on SM (CAD D. Vurpillot).


V. Surface comparison from smoothed models

To be continued…


Leave a Reply

Your email address will not be published. Required fields are marked *