Novel approaches for enhanced visualisation and recognition of rock carvings at Stonehenge

Abstract

The sarsen uprights at Stonehenge feature the largest panels of Early Bronze Age axe-head carvings in the world. Archaeologists use these carvings to better understand the significance of the monument. Between 2011 and 2012, the analysis of laser scanning and photogrammetric data led to the identification of 71 axe-head carvings and one dagger carving, in addition to the 44 carvings already known. Recent advances in carving visualisation and machine learning warrants a reanalysis of this data using new methods. Two novel techniques for carving visualisation, difference of Gaussians and pseudo-depth mapping, are introduced and compared to four recent techniques, radiance scaling, openness, distance between meshes, and extended difference of Gaussians. On the northwest face of Stone 53, difference of Gaussians highlighted the presence of two previously unidentified carvings, ten potential areas of carving, and nine alternative interpretations on previously found carvings. Pseudo-depth mapping revealed the presence of a further two previously unidentified carvings. In addition, an existing classifier for 3-D shape representation, MeshNet, is converted into a technique for carving recognition. MeshNet achieved 90.7 % accuracy on labelling samples of surfaces at Stonehenge with and without carvings, close to the benchmark performance of 91.9 % on ModelNet40. Both difference of Gaussians and pseudo-depth mapping can be implemented for visualisation of highly faded rock carvings in under two hours and under ten minutes respectively, while the application of MeshNet serves as a feasibility study of semi-automated carving recognition.