PAGE validator: respect reading/textline order/direction

[bertsky]

(in multi-level text consistency checks)

This takes existing ideas from ocrd_tesserocr.recognize.page_update_higher_textequiv_levels() into the validator (only slightly more general).

This should be considered a (late) follow-up on OCR-D/assets#16. In particular, one should verify the refined rules look better on real-world GT which contains (concatenated) subregions, non-default @textLineOrder and non-default @readingDirection, and then re-visit the existing textual consistency rules laid out in the spec.

[codecov-io]

Codecov Report

Merging #442 into master will decrease coverage by 2.09%.
The diff coverage is 59.30%.

@@            Coverage Diff             @@
##           master     #442      +/-   ##
==========================================
- Coverage   81.93%   79.83%   -2.10%     
==========================================
  Files          39       40       +1     
  Lines        2314     2440     +126     
  Branches      426      463      +37     
==========================================
+ Hits         1896     1948      +52     
- Misses        346      408      +62     
- Partials       72       84      +12     

Impacted Files	Coverage Δ
..._validators/ocrd_validators/workspace_validator.py	84.61% <ø> (ø)
ocrd_validators/ocrd_validators/page_validator.py	78.07% <59.30%> (-3.18%)	⬇️
ocrd_utils/ocrd_utils/logging.py	96.15% <0.00%> (-3.85%)	⬇️
ocrd_utils/ocrd_utils/__init__.py	66.01% <0.00%> (-0.92%)	⬇️
ocrd/ocrd/cli/workspace.py	71.79% <0.00%> (-0.36%)	⬇️
ocrd_utils/ocrd_utils/constants.py	100.00% <0.00%> (ø)
ocrd_models/ocrd_models/ocrd_page.py	100.00% <0.00%> (ø)
ocrd_models/ocrd_page_user_methods.py	0.00% <0.00%> (ø)
ocrd_models/ocrd_models/ocrd_mets.py	94.61% <0.00%> (+0.06%)	⬆️
ocrd/ocrd/processor/base.py	88.04% <0.00%> (+0.68%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d7f715f...fc31557. Read the comment docs.

[kba]

Looks great insofar as I can follow. I would really appreciate tests. We have some samples with ReadingOrder but they are very basic. Can we provide better test data? @tboenig

[bertsky]

@kba the last commit prevents an exception (instead of a negative result) during validation when coords have less than 4 points or baselines less than 2.

Looks great insofar as I can follow. I would really appreciate tests. We have some samples with ReadingOrder but they are very basic. Can we provide better test data? @tboenig

IIUC the main obstacle for GT activities is the lack of OCR-D/ocrd_segment#32 which itself is due to missing progress on #443. In other words: We need a way to fix most of the gazillions of GT errors now reported by the existing (already merged) coordinate validator automatically before we can move on.

[kba]

Looks good and PRImA-Research-Lab/PAGE-XML#23 shows the benefits. I will merge tomorrow unless @tboenig objects.

[kba]

@bertsky say the word and I'll merge :)

[bertsky]

say the word and I'll merge :)

I'd say it's not perfect now, esp. 6bf98d0 if PRImA-Research-Lab/PAGE-XML#23 does turn out to reveal PAGE coordinates are all "pixel-center" not "pixel-below-right" (in which case we would have a lot of work going over all processors) – but much better than before, anyway. So let's revisit when a verdict on coordinate semantics has been found (but merge for now).