PyPdfiumDocumentBackend wrongly merges text bonding boxes

[pedrolourencoribeiro]

Bug

PyPdfium2 produces very fragmented cells, with sub-word level boundaries, in many PDFs.
There is some logic to try to merge them

merged_text = "".join(cell.text for cell in group)
  merged_bbox = BoundingBox(
      l=min(cell.rect.to_bounding_box().l for cell in group),
      t=min(cell.rect.to_bounding_box().t for cell in group),
      r=max(cell.rect.to_bounding_box().r for cell in group),
      b=max(cell.rect.to_bounding_box().b for cell in group),
  )

While this works for most cases sometimes the bounding boxes overlap and we end up having duplicated characters. We could instead get the text_bounded based on the new computed BoundingBox

merged_bbox = BoundingBox(
    l=min(cell.rect.to_bounding_box().l for cell in group),
    t=min(cell.rect.to_bounding_box().t for cell in group),
    r=max(cell.rect.to_bounding_box().r for cell in group),
    b=max(cell.rect.to_bounding_box().b for cell in group),
)

bbox = merged_bbox.to_bottom_left_origin(page_size.height)
merged_text = self.text_page.get_text_bounded(*bbox.as_tuple())

loans-leases-with-images.pdf

...

Steps to reproduce

Parse the pdf provided (page 3 for example) and look for offffering

One example in the following PDF is
BBox1: off
BBox2: ffering
that leads to: offffering
...

Docling version

2.31
...

Python version

3.12
...

[pedrolourencoribeiro]

#1549
I've created a PR with the potential fix not sure on how to proceed