Improving ancient Roman coin classification by fusing exemplar-based classification and legend recognition

Abstract

In this paper we present an image-based classification method for ancient Roman Republican coins that uses multiple sources of information. Exemplar-based classification, which estimates the coins' visual similarity by means of a dense correspondence field, and lexicon-based legend recognition are unified to a common classification approach. Classification scores from both coin sides are further integrated to an overall score determining the final classification decision. Experiments carried out on a dataset of 60 different classes comprising 464 coin images show that the combination of methods leads to higher classification rate than using them separately. © 2013 Springer-Verlag.