A System for Processing and Recognition of Greek Byzantine and Post-Byzantine Documents.

Processing and recognition of Greek Byzantine and Post-Byzantine (old Greek) Documents has been proven to be a tedious task in the domain of Historical Document Image Processing. Several unique characteristics of these documents (existence of character ligatures, abbreviations, lack of clear word division, existence of symbols or punctuations in an arbitrary position) impose significant difficulties for current processing and recognition tools. In this work, we introduce a system for processing and recognition of old Greek documents and give details about all the components that comprise it. These include an image pre-processing, a text line segmentation and a recognition module. In order to test the proposed system, we introduce and provide publicly a new dataset of old Greek Documents that includes text line images and the corresponding transcription. Using this dataset, we evaluate the embedded recognition engine of the proposed system which is the open-source Calamari-OCR engine employing a variety of configurations. The best result corresponded to a character error rate less than 1.5% which is acceptable and promising. Finally, we also achieved promising results when comparing the embedded OCR engine with other recognition methods already proposed for the recognition of old Greek Documents.