Painting style-based recognition of potters: using convolutional neural network techniques

Abstract

This study explores and innovatively proposes a paradigm for applying Convolutional Neural Networks (CNN) to the micro-analysis of painted pottery production in archaeology. An ethnoarchaeological study of three modern painted pottery workshops reveals that the dot patterns painted by three different potters exhibit distinct structures and degrees of regularity, reflecting their unique painting styles. These stylistic differences are crucial for effectively distinguishing pottery painted by individual potters, and CNN techniques have proven highly effective in identifying potters with distinct styles. Further application of this technique to painted potteries from the second phase of the Miaodigou site demonstrates that the potteries can be categorised into at least three groups, each exhibiting a distinct painting style. This suggests that at least three potters (or three groups of potters) were involved in the production of the pottery, each displaying unique preferences in decorative motifs, overall composition, and stylistic execution.