Automated classification of petroglyphs

In this paper, we address the problem of automated petroglyph classification in a large real-world dataset. The dataset which contains more than 1000 petroglyphs is based on tracings from the UNESCO world heritage site Valcamonica, Italy and is expert-classified into two parallel typologies. For automated classifications of petroglyphs we utilise a combination of existing shape descriptors and a recently developed graph-based petroglyph descriptor. We achieve good classification results. We evaluate how the results can be incorporated into the daily work of archaeologists. We demonstrate that our tools can clearly enhance the process of manual classification.