Visualization and Automatic Typology Construction of Pottery Profiles
Laurens van der Maaten, Guus Lange, Paul Boon · 2009
Cultural Heritage Agency. Amersfoort, The Netherlands. Since the second half of the last century, computers became available to archaeologists. Since then, many have used various multivariate analysis techniques in the classification of archaeological objects with more or less success. Over the last decades, the popularity of multivariate analysis seems to have seized, partly because of the frequently disappointing performance of techniques such as principal components analysis and factor analysis. In particular, these techniques are hampered by their linear nature and by their emphasis on retaining global data structure. In this paper, we discuss a recently proposed technique for multivariate analysis, called t-SNE, which overcomes many of the weaknesses of principal components analysis and factor analysis. We illustrate the strong performance of t-SNE in experiments on a dataset of pottery profile drawings, in which we combine t-SNE with a shape matching technique based on shape contexts. Next to our discussion on t-SNE, we discuss a recently proposed affinity-based clustering technique, called affinity propagation. We show how affinity propagation can be used for the automatic construction of a typology of the shape profile drawings.