OCR-D processors: Propagate TextEquiv from line to region and page

[kba]

    Good idea to remove the words below. But how about text consistency upwards in the hierarchy (i.e. concatenation on region level)?

Originally posted by @bertsky in #9 (comment)