[Bug]: After updating from 17.1.0 to 17.2.0 marking text became difficult

[realsimix]

Describe the bug

With 17.1.0 when marking text it was highlighted in solid blue with the chars in white. Now with 17.2.0 the highlighting is in transparent light blue and no written text on it. Additionally the highlighted text and the text in the document often don't match well.

Is this new behavior wanted or do I do something wrong? I really liked the way text could be marked in 17.1.0.

Thanks for any insights.
Simon

Steps to reproduce

1. Run ocrmypdf --language deu+eng+fra+ita --rotate-pages --rotate-pages-threshold 5 \
--deskew --clean --output-type pdfa \
--sidecar file.txt file.jpg file.pdf
2. Open file.pdf
3. Mark text in the document.

Files

No response

How did you download and install the software?

No response

OCRmyPDF version

No response

Relevant log output

[jbarlow83]

What PDF viewer are you using? Can you take a few screenshots or video of behavior you're seeing?

There was a bug in 17.1.0 that caused rendering of OCR text. Normally an image was drawn on top so it wouldn't be seen, but if Tesseract also messed up on text positioning and if the image didn't cover the full page or was transparent, you'd see the text. Generally we don't want OCR text visible, but we do want visible on highlight.

However, there are other ways I might be able to get this to work....

[realsimix]

I have tested the PDF in Firefox now and indeed the text is not shown there, only with evince I see the text.

Unfortunately the created PDF is still better with 17.1.0 than with 17.3.0. I'll upload two PDFs so you can see it yourself. In the 17.1.0 file the word "Behandlung" is where it should be but in 17.3.0 it's not quite right.

Do you have an idea where it comes from?

17.1.0.pdf
17.3.0.pdf

[Kyrylo-Hrytsenko]

Hi @jbarlow83. May I ask if this is something that's planned to be fixed? I'm facing this issue as well.

What happens is that when words are far from each other (for example, in table cells, but not exclusively), the text gets stretched. I've attached examples of this behavior. Please note that for these screenshots, I intentionally deleted the image and kept only the text layer, making it visible to clearly show the stretching behavior.

Additional context:

• Not file-specific: This happens on almost all documents that have similar text gaps; it is not an isolated issue with one specific file.
• Workaround: This doesn't happen when using the sandwich renderer.
• Regression: This seems to be a regression, as the issue is absent in older versions (v17.10.0, v17.0.1, and v16.13.0).

Original doc: [OCR]S&P Global Ratings12133.pdf

OCRmyPDF v17.3.0
OCRmyPDF v17.1.0

[realsimix]

Is there a chance we can get a release soon with this fix applied? That would be great!

Thanks,
Simon

[jbarlow83]

There's a potential fix on main.
This kind of issue needs to be tested on a lot of platforms.