Improve the detection of justified text-frames. Usually these are not detected but several small frames instead. Also Frames, that are very close together, may create problems as lines from such frames being very close together, may be detected as one line. This whole problematic can be changed by an additional step in the frame detection, that create "heuristic" bounding boxes for sequences of lines and compare this with the already found frames.

examples please see the "problem" folders in the nextcloud