greek letters with diacritics appear as rectangles

[hubitor]

I'm testing OCRmyPDF on some greek documents. When I copy and paste the text from the OCRed file into a text editor all letters with diacritics appear as rectangles.

$ ocrmypdf -l ell -v 1 Ellada.pdf Ellada.ocr.pdf
DEBUG - ocrmypdf 5.2
DEBUG - tesseract 3.04.01
DEBUG - os.symlink(Ellada.pdf, /tmp/com.github.ocrmypdf.0xor195l/origin)

---

Tasks which will be run:

Task enters queue = 'ocrmypdf.pipeline.triage'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/origin, /tmp/com.github.ocrmypdf.0xor195l/origin.pdf)
Completed Task = 'ocrmypdf.pipeline.triage'
Task enters queue = 'ocrmypdf.pipeline.repair_pdf'
DEBUG - <PdfInfo('...'), page count=1>
Completed Task = 'ocrmypdf.pipeline.repair_pdf'
Task enters queue = 'ocrmypdf.pipeline.split_pages'
Task enters queue = 'ocrmypdf.pipeline.generate_postscript_stub'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.page.pdf, /tmp/com.github.ocrmypdf.0xor195l/000001.ocr.page.pdf)
Completed Task = 'ocrmypdf.pipeline.split_pages'
Task enters queue = 'ocrmypdf.pipeline.orient_page'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.ocr.page.pdf, /tmp/com.github.ocrmypdf.0xor195l/000001.ocr.oriented.pdf)
Completed Task = 'ocrmypdf.pipeline.generate_postscript_stub'
Completed Task = 'ocrmypdf.pipeline.orient_page'
Task enters queue = 'ocrmypdf.pipeline.rasterize_with_ghostscript'
Task enters queue = 'ocrmypdf.pipeline.skip_page'
Uptodate Task = 'ocrmypdf.pipeline.skip_page'

WARNING:
In Task 'ocrmypdf.pipeline.skip_page':
No jobs were run because no file names matched.
Please make sure that the regular expression is correctly specified.

DEBUG - Rasterize 000001.ocr.oriented.pdf with png16m
DEBUG -
Completed Task = 'ocrmypdf.pipeline.rasterize_with_ghostscript'
Task enters queue = 'ocrmypdf.pipeline.preprocess_remove_background'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.page.png, /tmp/com.github.ocrmypdf.0xor195l/000001.pp-background.png)
Completed Task = 'ocrmypdf.pipeline.preprocess_remove_background'
Task enters queue = 'ocrmypdf.pipeline.preprocess_deskew'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.pp-background.png, /tmp/com.github.ocrmypdf.0xor195l/000001.pp-deskew.png)
Completed Task = 'ocrmypdf.pipeline.preprocess_deskew'
Task enters queue = 'ocrmypdf.pipeline.preprocess_clean'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.pp-deskew.png, /tmp/com.github.ocrmypdf.0xor195l/000001.pp-clean.png)
Completed Task = 'ocrmypdf.pipeline.preprocess_clean'
Task enters queue = 'ocrmypdf.pipeline.select_ocr_image'
Task enters queue = 'ocrmypdf.pipeline.select_visible_page_image'
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.pp-clean.png, /tmp/com.github.ocrmypdf.0xor195l/000001.ocr.png)
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.page.png, /tmp/com.github.ocrmypdf.0xor195l/000001.image)
Completed Task = 'ocrmypdf.pipeline.select_ocr_image'
Task enters queue = 'ocrmypdf.pipeline.ocr_tesseract_hocr'
DEBUG - ['tesseract', '-l', 'ell', '/tmp/com.github.ocrmypdf.0xor195l/000001.ocr.png', '/tmp/com.github.ocrmypdf.0xor195l/000001', 'hocr', 'txt']
Completed Task = 'ocrmypdf.pipeline.select_visible_page_image'
Task enters queue = 'ocrmypdf.pipeline.select_image_layer'
DEBUG - 1: page eligible for lossless reconstruction
DEBUG - os.symlink(/tmp/com.github.ocrmypdf.0xor195l/000001.ocr.oriented.pdf, /tmp/com.github.ocrmypdf.0xor195l/000001.image-layer.pdf)
Completed Task = 'ocrmypdf.pipeline.select_image_layer'
Completed Task = 'ocrmypdf.pipeline.ocr_tesseract_hocr'
Task enters queue = 'ocrmypdf.pipeline.render_hocr_page'
Completed Task = 'ocrmypdf.pipeline.render_hocr_page'
Task enters queue = 'ocrmypdf.pipeline.combine_layers'
Completed Task = 'ocrmypdf.pipeline.combine_layers'
Task enters queue = 'ocrmypdf.pipeline.merge_pages_ghostscript'
DEBUG - Final pages: /tmp/com.github.ocrmypdf.0xor195l/000001.rendered.pdf
/tmp/com.github.ocrmypdf.0xor195l/pdfa.ps
DEBUG -
Completed Task = 'ocrmypdf.pipeline.merge_pages_ghostscript'
Task enters queue = 'ocrmypdf.pipeline.copy_final'
Completed Task = 'ocrmypdf.pipeline.copy_final'
INFO - Output file is a PDF/A-2B (as expected)
DEBUG - <PdfInfo('...'), page count=1>

Test file (excerpt from Wikipedia):
Ellada.ocr.pdf
Ellada.pdf

[jbarlow83]

I recommend upgrading to Tesseract 3.05 which should resolve this issue. (Tested by me.)

If you prefer to not upgrade, you could try the following (not tested):

ocrmypdf --pdf-renderer tesseract --output-type pdf -l ell <input> <output>

which will select a PDF renderer that tends to preserve less of the PDF formatting but does better on non-Latin scripts.

[jbarlow83]

v5.3 now gives a warning in this case

[hubitor]

Thanks for the quick response! Just upgraded to tesseract 4 and tested on the same file. Not only there are no rectangles but the result is almost perfect!