Automatic Classification of Ceramic Sherds

The ARCADIA project focuses on methods to facilitate the interpretation and the evaluation of an archaeological ceramic heritage. It consists in developing a digitalization chain of the ceramic sherds to automatically extract and classify the patterns imprinted by potters using a carved wood wheel. In this paper, we present a pattern extraction and classification method that aims to solve these issues. Encrusted patterns are extracted from 3D scans using various techniques and are then classified using a Support Vector Machine (SVM) model. Results are first evaluated in terms of patterns extraction and then the classification step is compared to the ground truth obtained manually (inking patterns performed by the archaeologist). Keywords— Archaeological ceramic heritage, Carved wood wheel decoration, Classification, Pattern recognition.