(finding polygonal areas) We provide tools for finding separations among rectangular regions, but not at this stage. Instead we wait until after skew detection. If your processing requires that shapes other than rectangles be found, we do not support this.
We also assume (see just below) that rectangles filled with text will occur aligned to vertical and horizontal axes, and that breaking up of the text can be done by alternate vertical and horizontal cuts. This is not always true (for example, ``sidebars'' embedded half-way in partial columns violate this heuristic.) There are heuristic methods described in the literature that generally find such situations, and they are not inconsistent with our representation; it hasn't struck us as important.