Character recognition for Greek squeezes
摘要
Squeezes are three-dimensional paper impressions made of stone inscriptions, a form of historical document where digital processing methods have received relatively little study to date. This paper reports on experiments in text recognition, working with a collection of approximately 30,000 squeezes originally collected from museums and archaeological sites in classical Greece. It explores a number of complementary strategies for character recognition in this medium, with the aim of establishing a strong benchmark in performance. Based on these experiments, we identify a recognition pipeline based on scene text recognition algorithms combined with line-based recognition that achieves a character error rate around 13%.