Now Apostolos Antonacapoulos of USAL is presenting Permormance Evaluation in Document Recognition Workflows, showing the way the sophisticated Ground Truth in IMPACT is produced, and how it is used to evaluate all tools that contibute to OCR results. He tells us that for him, as an image scientist, IMPACT was a wonderful opportunity to gather real world datasets form the libraries to try out their work.
Before him, two service providers presented their work: Michael Fuchs from ABBYY Europe showed the improvements in a.o. binarisation for the currently available FR 10 OCR engine in IMPACT, and the promising work done on the integration of the dictionaries for historic texts (there are to be nine of these in IMPACT eventually!) in the FineReader.
Jean-Pierre Raysz of Jouve , market leader in e-publishing cultural heritage, talked about the management of OCR digital libraries, showing impressive targets for accuracy (up to 99,9999%) in digitized content.
On the other end of the research spectrum, Kamel Ait Mohand Litis showed his work on OCR adaptation of characters for old documents, an interesting approach that has already proved some results.