PDF renderer: Tesseract inserts spaces for non-text blocks it finds

[bleze]

Environment

• Tesseract Version: 5.2.0 and 4.1.3
• Platform: Windows 10, x64

Current Behavior:

Spaces are found by Tesseract and inserted in output PDF. It seems to be confused by the horizontal lines in the attached image.
I use this PDF to overlay on top of another PDF which already contains text, and this causes problems in the resulting PDF due to intersections.

Expected Behavior:

Tesseract finding standalone spaces does not make sense to me. I would expect Tesseract to only find characters in the logo and disregard the lines. At least the lines should be underscores or something - not spaces.

Suggested Fix:

Unsure if problem should be fixed in Tesseract or if the spaces should be filtered in the PDF renderer. I think Tesseract as it will fix all output formats.

[amitdo]

Please post hocr output.

[bleze]

hocr file created like this; tesseract.exe "C:\Tesseract\v.png" "c:\tesseract\v" --tessdata-dir "C:\Tesseract\tessdata_best-main" -l dan --psm 4 --oem 1 -c tessedit_create_hocr=1

Let me know if these parameters at not correct, thanks.

v.zip

[amitdo]

Tesseract see this as a photo + a few line separators.

The hocr renderer correctly drops the spaces.

The pdf renderer should ignore any non-text blocks.

[amitdo]

Related PR: #3723.

[amitdo]

@bleze,

See the fixed pdf output here:
#3959 (comment)