We were working on testing the possibilities of image base modelling and photoscan to capture the geometry and colour of glass pieces. The idea was to create models of shards that can be later used to make 3ds max reconstructions of the whole vessels.
The material came from the excavations at Iron age setellment site at Uppåkra. The nature of the galss finds from the setellment sites makes them dificult to identify, as the shards are very fragmentetd and the material can be disspresed around the features, making it hard to connected the fragments belonging to one vessel. We were aware of the possible limitations, and the outcome of the first week of our work can hardly be called satisfying, but we managed to make some useful observations and developed some ideas about the possible solutions to many problems we encountered. Prelininary results can be considered as promissing, but more reaserch will be requiered.
Acquisition:
Glass is probably one of the most demanding materials when it comes to image based modelling. It is reflective and transparent at the same time. Experiments with modern examples, outside the studio enviroment, were a series of not very satisfying events, but fortunately the real artefacts display a bit different qualities, being more opaque and having more distinct features on the surface that could be captured by camera. We knew that this will be a though experience.
Unfortunately due to the holiday season and time constraints we could conduct only one campaign at the museum storehouse. We have selected initially three pieces that displayed different qualities:
 |
A transparent claw beaker fragment. The shard has probably been a part of a rim of a beaker
|
|
A transparent claw beaker fragment. The shard has probably been a part of a rim of a beaker
|
 |
A burnt shard, most likely from a Snartemo beaker.
|
 |
A piece of green glass with polished ovals. This shard is about 0.9 cm thick.
|
 |
| backside of the polished glass bowl |
We used two defused lights on the stands to illuminate the white background on which the artefact was placed. The lights were mostly directed on the walls and the celling of the photo room to provide even more diffused conditions. We were using Sony alfa300 camera with sigma 18-200 mm lens and an IPhone 4. We run some test using the tripod that was available at the storehouse. We tried to process the images at the spot to check the quality of the acquisition. Surprisingly best results were reached with IPhone used as a ´free camera’. The Sony camera was focusing too much on different spots on the glass instead of the whole pieces and that resulted in the less dense point clouds. The results were better when a simple grid made out of lined piece of paper with some symbols drawn on it was used.
Post processing in Photoscan and Photoshop.
We uploaded the photos from different cameras and different acquisitions. Initially we started processing them just straight “of the memory card”. Soon we realized that better effect could be reached if the images were first edited in Photoshop. We used it mainly to boost the contrast, reduce the brightness and change colour a bit to improve the initial point detection outcome.
 |
The back side of the burned piece after photoshop.
|
 |
The front side of the burned piece after Photshop.
|
Here the .raw format that was used by high quality images from Sony alfa300 offered a lot of different possibilities, unfortunately the time did not allowed us to test everything. Also altering the colour might be considered a problem, since it is an important quality in the research related to ancient glass. Other problematic fact was the scale of the shards - when confronted with small objects placed on a flat surface Photoscan creates a lot of artificial bumps on it, that t mainly occlude the edges, what is a major setback in the next step of the process.
 |
| The Photoscan model without texture, from the IPhone acquistion |
 |
The Photoscan model in a wireframe visualization
|
 |
| Photoscan model with texture |
The output was quite satisfying. The surface was a bit bumpy, probably due to the reflective glass surface and the top and bottom halves of shards had yet to be connected.
Cleaning of the models and alignment in MeshLab
To clean and align the shards we used MeshLab. It was a difficult process since the only part that was common for both models were the edges, that usually are only couple of millimetres thick, and they get easily distorted in Photoscan. The way to solve that problem would probably be to use the rough manual glue tool, but it would be a long a difficult process. Other problem that we encountered was connected to the fact that we would lose all the textures if we merged parts of the model. The solution was to re-project the colour information from high quality textures to the vertexes and later use sampling to colour the poison filter model made out of both merged meshes. This simple process however was impossible to complete since there was a problem with mesh flattening - programed crashed each time we run it, despite the fact that most of cleaning and repairing filters were used before the process. Meshlab offers however excellent filters that can be used to smooth out the mesh, for example Laplacian smooth.
 |
| roughly cleaned photoscan model without texture |
 |
| Photoscan model with texture |
 |
| smoothed photoscan model without texture |
 |
| smoothed photoscan model with texture |
Preparing the reconstructions and placing the pieces in 3ds max
After creating models in Meshlab we needed to have some reconstructed vessels to place them on for presentation. These were made in 3ds max using drawings and photos as the source materials. We mainly use the lathe modifier to create the “body “ of the glass vessel to which additionally modelled parts - like strings made using sweep modifier or claws were added. Modelling could be difficult since it is a challenge to place the strings directly on a bend surface of the beaker.
The hardest part was however to achieve a good effect by working with v-ray material. It is a powerful tool that offers many possibilities but the ancient glass in a very hard to model, since each vessel displays a bit different qualities. There are many levels of transparency, different colours and different scales of reflections. It is possible to spend hours on working to achieve a perfect outcome. In our case the time was however a limiting factor. We managed to produce some renderings, but to save time just the front parts of the modelled pieces were placed on reconstructions.
 |
The recontruction was made with this drawing of a bowl with polished ovals and the Uppåkra shard, as a reference.
Drawing from Stjernquist 2004. |
 |
Reconstruction, in 3d studio max, of a bowl with polished ovals. The Uppåkra shard could have been a part of a bowl of this type.
 |
 |
The model of the shard attached to the reconstructed bowl.
|
 |
| The reconstruction of the Snartemo beaker has been made with this drawing and the Uppåkra shard, as references. Drawing from Näsman 1990 |
 |
Reconstruction of a beaker of Snartemo type. These beakers can have different shapes and colours so this is one interpretation made with the shard and a drawing as reference.
|
Conclusion and further perspectives
There is still more reaserch needed when it comes to application of imaged based modeling to the glass reconstruction. Our results are promissing, but a more standarised method and approach must be developed. More test are needed, especially with the aquistion techniques. A bigger sample of glass pieces is needed to be processed.
Probably using a powder on pieces and scanning them by the laser scanner with the later reprojecton of color from photos might have been an easier solution. It is possible to create a model and reconstruction starting from image based modeling. We are still not sure if it can be applied to every piece, or will it only work on selected examples, displaing certain qualities, such as small transparency and high level opaquness. It is a difficult process that probably gets much easier with experience, and more impressive results can be reached in a longer term perspective. A training in light setting and studio photography could probably influence the results, since the aquistion part is always the cruicial link in working with image based modeling.
Amanda and Franciszek
References
Näsman, U. 1990. Om fjärrhandel i sydskandinaviens yngre järnålder. Handel med glas under germansk järnålder och vikingatid. Hikuin 16, 1990, pp. 89-118.
