Data-driven Automatic Attribution of Azerbaijani Flat Woven Carpets

Abstract

Carpet attribution is an important task for studying the carpets and textiles, and more generally the history and culture of the communities producing these carpets. However, this is not an easy task, often relying on experts' subjective opinion or complex chemical or radiographical analysis, often not available to many practitioners. In this work, building on the success of applying machine learning and artificial intelligence methods in different fields, we present another, data-driven approach for carpet attribution. Based on a large dataset of Azerbaijani flat woven carpets we have developed a novel machine learning based data-driven carpet attribution system, which successfully determines their types, schools and weaving century, achieving up to 98% accuracy of the attribution.