Images and grouth truth data for the blog post https://comsys.informatik.uni-kiel.de/lang/de/res/optimizing-binarization-for-ocropus/
You need OCRopus https://github.com/tmbdev/ocropy and my character model for the Austrian-Hungarian casualty lists https://github.com/jze/ocropus-model_oesterreich-ungarn
I used version f7b163c of OCRopus and version b095693 of the character model.
The script run.sh will perform all steps to measure the character recognition error rate. It expects that OCRopus' programs are executable (in $PATH) and that the character model is located in a sibling directory to this directory.