Modelling the Headda Stone, Peterborough Cathedral

The Headda Stone in its current position in Peterborough Cathedral
The Headda Stone in its current position in Peterborough Cathedral

The Headda Stone is a remarkable and unique example of a Middle-Saxon sculpted stone, perhaps the focus for a shrine, in the form of a ‘house’ with carvings of Christ, the Virgin and 10 other figures (presumably 10 of the 12 apostles), set within an arched colonnade, with 6 figures on each of the long sides of the stone. The whole block measures a little more than 1m x 0.75m x 0.35m and has been weathered and worn to varying degrees. The ends appear to have been reworked not only removing carved detail but at one end adding a date of 870, considered by Prof. Rosemary Cramp and others to be considerably later than the true date of the sculpture. The ‘roof’ is decorated with four ornamental panels on each slope.

The Headda stone currently sits in the Lady Chapel behind the High Altar in Peterborough Cathedral but is thought likely to have come from the former Anglo-Saxon monastic house that preceded the Norman Cathedral. The stone, in its present position is clearly visible with the side showing Christ, the Virgin and 4 other figures facing out into the chapel; but the other side is very difficult to view as is sits a few centimetres away from curved wrought iron railings at the back of the Lady chapel.

Creating a visual record of the stone which can both enhance our ability to interpret the eroded sculpture and make it ‘virtually’ available both for researchers and the general public on the internet is particularly challenging. Over the last 15 years the ever increasing power of our computers and related scientific and survey instruments has enabled us to record three dimensional surfaces in ways that could not have been predicted 25 years ago.

Laser scanners, which use the time of flight of reflected laser beams to create very high density 3D point-clouds comprising many millions of reflected points, can be used to create highly accurate digital 3D models which can be viewed on high powered desktop computers and on the internet. Laser scanners remain very expensive, require the use of complex and also expensive software and powerful hardware as well as highly trained operators and a lot of time to process the results. These factors and the frequently inconvenient locations of many ecclesiastical sculptures reduce the obvious potential of these instruments for practical and simple operation for recording the magnificent assemblage of early medieval carved stone found in churches and cathedrals in Britain. Over the last few years alternative approaches have been developed that allow us to generate high precision 3D computer models using affordable digital cameras, computers and easy to use software.

The Headda Stone seen from the 'back' obscured by railings
The Headda Stone seen from the ‘back’ obscured by railings

3Di or 3D imaging using digital photogrammetry is a relatively new technique for creating digital 3D models  from digital photographs at everything from the landscape scale to individual small objects. 3Di models are generated using software that employs sophisticated algorithms including Structure From Motion and  Bundle Adjustment which, given the right number and quality of overlapping digital photographs can calculate the relative position of the camera used to gather the images and then, using combination of camera positions and already identified image pixels further calculate the 3D position of the pixels in each image With this information the software can construct an accurate digital 3D model of the imaged surface for viewing and further analysis on a desktop computer or distribution on the internet. The 3d models are represented initially as point-clouds of multi-million individual 3D points coloured according to the original 2D photographs. A triangular network or mesh is calculated which maps to the surface of the point-cloud using varying sizes of triangles according degree of undulation in the surface, over which texture or fragments of the original digital images are applied to give the impression of 3D texture.

The modelling of the Headda stone posed a particular challenge on account of the presence of the railings which extend across the back of the stone, in an area which also has poor lighting made more problematic by higher levels of lighting coming from the windows behind the stone. The Headda stone modelling exercise was undertaken for multiple reasons; to test the viability of 3Di to accurately record the form of this sculpture despite the difficult location, to demonstrate that 3Di offered the potential to greatly enhance the record of ancient stone sculpture, to create a resource that could be shared in 3D on the internet and by mobile phone and produce an accurate archive that would support and enhance the archives of the Corpus of Anglo-Saxon Stone Sculpture, published and maintained through the University of Durham, and contribute to the curation and interpretation of the object by the Dean and Chapter at Peterborough Cathedral.

3Di image generation from digital photographs requires high quality, well focussed and sharp photographs. In this case an Olympus OM-D EM-1 and EM-10 camera bodies, recording 16 megapixel images using an Olympus M.Zuiko 17mm. f1.8 lens ( many thanks are due to Olympus UK for the loan of this lens). The standard zoom lenses sold with most digital cameras are rarely as sharp as single focal length ‘prime’ lenses or professional zooms. An LED light panel was mounted upon the flash hot-shoe on the camera, and the camera mounted on a solid tripod.

Positioning the camera to photograph the stone through the railings
Positioning the camera to photograph the stone through the railings

In order to create the 3Di model, images which cover the whole of the surface to be modelled must be photographed with 60% overlap across images in all directions, ideally the detailed observed in each pixel should be visible in minimum of 3 images taken from different positions. For the front of the stone which is not obscured it was easy to record a sequence of images across the stone at different heights, with additional images recorded going round the corners of the stone. The reveres of the stone required a more complex approach with images collected in a series of arcs ensuring that all parts of the surface were visible from different angles observed through the railings.

All photographs were exposed using manual exposure/aperture settings and recorded in JPEG and RAW formats. JPEG images which employ averaging algorithms to reduce the file size are not ideal for precision modelling, RAW images which store the full image details as recorded by the camera sensor were used for modelling. The RAW images were initially processed using DXO Optics Pro, software which uses specifications calculated for each lens and sensor combination to fine tune each image reducing distortion and increasing sharpness by tiny amounts. The processed files were then saved in TIFF format prior to use for modelling.

The 3D modelling software used for this project is Agisoft Photoscan Pro, selected from many different available packages following an in depth assessment of most of the available Structure from Motion (SfM) packages available in 2013/2014.   The data was processed using a very high specification PC running windows 7 64 bit on an i7 quad core processor with 48Gb of RAM and an NVIDIA GTX 670 graphics card with 2GB of graphics RAM.. Agisoft Photoscan Pro continues to lead the field in SfM processing although faster computing platforms can inevitably be found today.

Once the images had been checked for sharpness, a total of 345 images were selected, from a total of more than 800, for modelling; this is more than would normally be needed for optimum results, the railings however meant than the back of the stone needed many more images than would otherwise have been required.

Agisoft screen showing the calculated high density point-cloud including the railings
Agisoft screen showing the calculated high density point-cloud including the railings

Figure 4: Agisoft screen showing the calculated high density point-cloud including the railings

As a first attempt a model was generated without any attempt to remove the railings, this first model took about 48 hours to process completely; generating the 3Di model requires a number of different processing stages, calculating camera positions, generating a high density point cloud, a triangulated surface mesh and finally applying image texture to the mesh.

Oblique view of the point-cloud showing the modelled railings
Oblique view of the point-cloud showing the modelled railings and camera positions

The point cloud comprised about 63 million points, from which meshes of different densities were required for archive, high and low resolution models for internet display purposes. Had it not been for the complexity of recording ‘through’ the railings far fewer images could have been used, and the model may potentially be improved slightly visually by selectively removing any unnecessary photographs.

Camera positions around the Headda Stone seen from above
Camera positions around the Headda Stone seen from above

Agisoft provides mechanisms for constraining the volume to be fully processed and thus the model could be further processed by eliminating the railings from the model before final processing. The high density  point cloud produced by the software was very good with sub millimetre precision, but when the image texture was mapped onto the triangulated surface the texture included poorly defined stripes resulting from the presence of the railings. The only way to successfully eliminate the railings completely from the texture was to mask the images using a mask function in Agisoft to mask each individual railing and any shadow or merging of the pixels on the railing edges. This process was time consuming but in the end meant that all trace of the railings could be eliminated.

Headda Stone, masking the railings in Agisoft
Headda Stone, masking the railings in Agisoft

Once fully processed attention was turned towards producing output files that would be useful at a variety of different scales to facilitate viewing for different purposes and supporting different levels of computing power. The 3Di files themselves can be very large and require specialist software which in effect makes it difficult for interested non computer experts to use, however a new internet based software utility combining the functions of online storage and online display has been developed by a French software company Sketchfab, founded in 2012.

Agisoft and a number of other professional software products include functions that directly support export of files to Sketchfab, which not only provides a central repository for 3Di models but allows them to be enhanced though shading and viewed on almost any computer using the internet for file delivery. Some files will be too big for viewing on some computers and so files at multiple resolutions have been created, perhaps the most extraordinary reflection of current day computing is the fact that these models can also be viewed using a smart-phone.

In addition to creating 3Di files that could be viewed from anywhere using the internet researchers need results that can be measured and shared with colleagues and not necessarily on the internet. For this purpose 3D Adobe PDF files were generated with correctly scaled models using scaling information in a set of photographs that were use to scale the model during processing but omitted in the texturing of the final model. The 3D PDF format is an internationally accepted archive format which supports 3D measurements and can be viewed on most modern computers.

In addition to the textured models placed in the Sketchfab environment or saved in PDF files, a non textured high resolution model without the texture or colour information was also placed on Sketchfab as a monochrome triangulated model which is more suited to observing fine details of the surface form.

The 3Di modelling of the Headda Stone has been a great success and in addition to creating an easily accessible archive of 3D models, has revealed details that cannot be seen with the naked eye (eg: the figure on the right hand end of the reverse of the stone. The youthful face that emerges probably identifies the figure as either John, Thomas or Philip).

Headda Stone triangular mesh - it is this mesh that describes the shape of the surface and it is optimised to show more detail on curved rather than flat surfaces.
Headda Stone triangular mesh – it is this mesh that describes the shape of the surface and it is optimised to show more detail on curved rather than flat surfaces.
The final high density point cloud (more than 30 million points) generated by Agisoft Photoscan Pro
The final high density point cloud (more than 30 million points) generated by Agisoft Photoscan Pro
The high resolution on-line image viewed using Sketchfab
The high resolution on-line image viewed using Sketchfab

The final model and others generated as part of the same research project can be viewed using on the Sketchfab archive site in a folder entitled Soke of Peterborough.

With thanks to:

Prof. Jo Story, School of History (University of Leicester)

Dr Jackie Hall, Cathedral Archaeologist at Peterborough Cathedral

The Dean and Chapter of Peterborough Cathedral

Dr Sarah Semple, Department of Archaeology (Durham University)

Prof. Philip Dixon, Cathedral Archaeologist at Lincoln, Ely, and Southwell

 

Soke of Peterborough by d.powlesland on Sketchfab

 

Leave a Reply

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

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>