Image based Classification of Ancient Coins

Numismatics deals with various historical aspects of the phenomenon Money. Fundamental part of a numismatist`s work is the classification of coins according to standard reference books. Reference numbers make the full description of a given coin type (including accurate dating, the distinction between minting places or any available political background) obtainable for everyone. The classification of ancient coins is a highly complex task that requires years of experience in the entire field of numismatics. Computer Vision explores the theory and technology to obtain and interpret information from images. For the application of ancient coin classification, computer vision techniques like Symbol Recognition and Optical Character Recognition (OCR) are investigated. The aim of the proposed project is to develop a framework for the automatic image-based classification of ancient coins. The framework comprises an image acquisition step, where optimal conditions for the acquisition of coins are examined. Classification is achieved by extracting and matching discriminative features like local image features and coin inscriptions obtained by optical character recognition (OCR) from the images. As the project`s topic is interdisciplinary, it brings together competencies from the fields of computer vision and numismatics: the Pattern Recognition and Image Processing Group at the Vienna University of Technology and the Department of Coins and Medals at the Museum of Fine Arts, Vienna. The project`s basic research lies in the field of computer vision but has the goal to produce an application for automatic image-based classification of historical coins in large-scale databases in the long run. Therefore, it establishes a link between basic scientific research and the development of an innovative application in the future.

ILAC was an interdisciplinary project of numismatists from the Museum of Fine Arts in Vienna and researchers from the Computer Vision Lab of the Vienna University of Technology. Its goal was to develop computer vision methods for the automatic classification of ancient coins. For this purpose, the complete collection of the gold- and silver-coinage of the Roman Republican age (approx. 211 - 27 BC) was photographed and indexed and served as evaluation data for the development of the computer vision methods.For computer vision research, ancient coins are a challenging type of objects due to the high number of classes with high variability. For instance, for the investigated domain 550 coin types are defined. As a consequence of their non-industrial manufacturing, age and metallic relief-like structures various aspects such as non-rigid deformations, missing information due to abrasions as well as effects of illumination change have to be considered. These individual challenges have been tackled in the project by means of new, innovative algorithms. Finally, the algorithms were integrated into a robust image comparison method which allows to find the most similar coin image among a dataset of reference coin images.In addition, the extraction of semantic information for an improved classification performance was addressed. Legend recognition was established by means of a lexicon of known legend words. Furthermore, pre-learned image recognition models were utilized to detect common motives such as chariots or she-wolves on the coins. These additional information sources allow for a pre-selection of candidate coin types in order to improve both the classification accuracy and runtime of the method.The basic research conducted in this project has the potential to support and improve numismatic research in the near future. For instance, the developed methods can be used for a faster classification of coins or the automatic grouping of coin hoards. In the area of computer vision the conducted research contributes to more robust image comparison methods which can be further applied to a wider class of problems. In addition to the published scientific papers and conference presentations, a showcase featuring the imagery on roman-republican coinage was created at the KHM and on display from Nov. 2013 to June 2014. At the Lange Nacht der Forschung on April 4th, 2014, the project was presented, both its technological as well as its iconographical aspects.