Cuneiform Character Similarity Using Graph Representations

Abstract

Motivated by the increased demand for computerized analysis of documents within the Digital Humanities we are developing algorithms for cuneiform tablets, which contain the oldest handwritten script used for more than three millennia. These tablets are typically found in the Middle East and contain a total amount of written words comparable to all documents in Latin or ancient Greek. In previous work we have shown how to extract vector drawings from 3D-models similar to those manually drawn over digital photographs. Both types of drawings share the Scalable Vector Graphic (SVG) format representing the cuneiform characters as splines. These splines are transformed into a graph representation and extend these by triangulation. Based on graph kernel methods we show a similarity metric for cuneiform characters, which have higher degrees of freedom than handwriting with ink on paper. An evaluation of the precision and recall of our proposed approach is shown and compared to well-known methods for processing handwriting. Finally a summary and an outlook are given.