Building and Improving an OCR Classifier for Republican Chinese Newspaper Text

In our paper we present the first results from a systematic approach to full text extraction from a Republican China newspaper. Our basis is a small corpus for which also a ground truth exists. We present our character segmentation method which produces about 70.000 images of characters. Based on the hypothesis that pre-training on extensive amounts of suitably augmented character images will increase the OCR accuracy for evaluation on real-life character image data, we generated additional synthetic training data. We than compare the OCR recognition results and show that a combination of synthetic and real characters produces the best results. Finally, we propose a method that makes use of a masked language model to for OCR error correction.

Ein Beitrag zur 8. Tagung des Verbands "Digital Humanities im deutschsprachigen Raum" - DHd 2022 Kulturen des digitalen Gedächtnisses.