hocr-check complains assert doc.xpath("//meta[@name='ocr-id']")!=[]

[CharlesNepote]

Can be reproduced with both tesseract and gImageReader hOCR files.
manisandro/gImageReader#101

Does the script end with this error or is it still checking the other issues?

[kba]

I would recommend you to use hocr-spec-python. It's an explicit replacement for hocr-check which is more or less nonsensical at the moment.

[kba]

You can try it online here: http://digi.bib.uni-mannheim.de/ocr-fileformat/#validate

[CharlesNepote]

Every file I try to upload to hocr-spec-python online -- http://digi.bib.uni-mannheim.de/ocr-fileformat/#validate -- ends with "NameError: global name 'KeyErrora' is not defined" (from tesseract or gImageReader).

[kba]

KeyErrora is a typo in https://github.com/kba/hocr-spec-python/blob/dc6377b8b46d1cb81a8571ad71c22746758bb1a2/hocr_spec/spec.py#L410. I'll fix it, thanks.

[kba]

Should be fixed and deployed. BTW: If you have any samples, gladly complicated ones, you're welcome to contribute them to ocr-fileformat-samples.

[zuphilip]

However, the main problem here we should also fix in hocr-check:

• ocr-id should be replaced with ocr-system in the check itself and the tests
• Then we are again conform with the specs.
• Also current Ocropus outputs looks like this: https://github.com/kba/ocr-fileformat-samples/blob/master/samples/hocr/1.1/433934212_0017.html

I can do this change, but just want to make sure, that we agree here.

[kba]

Before you implement this further, I have a branch somewhere where I've done that, let me check.

Several actually.

[kba]

Ah, no I did not actually fix it, so yeah: ocr-id should be ocr-system. I would probably remove the containment checks and concentrate on the bounding box overlap check since that is actually an interesting check. And I don't think it's in the specs BTW.

Should we maybe use hocr-spec-python as a base to reorganize this (#42)? The parsing of title= attributes could come in handy.

[CharlesNepote]

I agree that ocr-id should be ocr-system.

[zuphilip]

I created a simple fix in PR #61 . We can also improve on that the ocr-check but maybe we should then open another issue for that to discuss further. I don't understand your comment, @kba, completely and I am not sure if I agree...

[kba]

Checking whether ocr_line is within ocr_page does not tell you much. Of course it should be checked but it's rather minor and only checking three classes seems arbitrary.

What's more important is that ocr_column is not in the spec anymore, this should be ocr_carea.