Logo Logo
Switch Language to German
Altaweel, Mark; Squitieri, Andrea (2020): Quantifying object similarity: Applying locality sensitive hashing for comparing material culture. In: Journal of Archaeological Sciencs, Vol. 123, 105257
Full text not available from 'Open Access LMU'.


We present a novel technique that compares and quantifies images used here to compare similarities between material cultures. This method is based on locality sensitive hashing (LSH), which uses a relatively fast and flexible algorithm to compare image data and determine their level of similarity. This technique is applied to a dataset of sculpture faces from the Aegean, Anatolia, Cyprus, Egypt, Iran, Indus/Gandhara, the Levant, and Mesopotamia. Results indicate that the objects can be differentiated based on regional differences and show similarities to other locations that share specific material culture traits. Images from known locations enable a network of compared objects to be constructed, where inverse closeness centrality and link weights are used to indicate areas that have a greater or less cultural similarity to other regions. Different periods are assessed, and the results demonstrate that objects from earlier than the 9th century BCE show greater similarity to other local and Egyptian items. Objects from between the 9th and 4th centuries BCE increasingly show interregional similarity, with the eastern Mediterranean, including the Aegean, Anatolia, Egypt, and Cyprus, having close similarity to multiple regions. After the 4th century BCE, greater sculptural similarity is found across a wide area, including the Aegean, Cyprus, Egypt, Mesopotamia, and Gandhara. In general, sculptures from more distant areas increase in similarity in later periods, that is starting from the 9th century BCE. The results demonstrate that the technique can be applied to quantifying object similarity and extended to a broad range of archaeological objects, while also being a tool for rapid analysis that requires minimal data compared to some machine learning techniques. The code and data are provided as part of the outputs.