Tesseract: "a" and "o" vowels replaced by unicode ordinal indicators for Portuguese language in the output

[scardine]

Environment

• Tesseract Version: 4.0
• Commit Number: 7afa05a03ed87a95a46da798fd11cae60f390441
• Platform: Ubuntu 17.04

Current Behavior:

For the following source images, the "a" and "o" vowels are replaced by "ª" and "º" (not the the superscript lower-case "a" and "o" but the feminine and masculine ordinal indicators, U+00AA and U+00BA) in the output:

$ tesseract notebooks/tag2.png stdout -l por
Declªrªç㺠de Nªscidº Vivº

Expected Behavior:

"a" and "o" verbatim:

$ tesseract notebooks/tag2.png stdout -l por
Declaração de Nascido Vivo

What I have tried:

• everything in the ImproveQuality wiki page
• changing page segmentation modes and OCR engine
• messing with character whitelists and blacklists
• custom user word and pattern files
• different fonts

Nothing made any difference.

Previous versions:

With 3.04 the output is like the expected.

[tomascharad]

Hi!

Same thing happens for spanish with letter 'o'

Original image
http://recursostic.educacion.es/observatorio/web/images/upload/1observatorio/iconos_art/texto.jpg

[scardine]

Confirmed, the same happens with the "o" in Spanish.

[amitdo]

https://en.wikipedia.org/wiki/Ordinal_indicator

[amitdo]

Can't reproduce this issue.

tesseract por2.png por2 -l best/por

Or

tesseract por2.png por2 -l best/por --oem 1

Output:

Declaração de Nascido Vivo

[amitdo]

https://github.com/tesseract-ocr/tesseract/wiki/Data-Files#updated-data-files-for-version-400

[scardine]

Nice. I was using pre-built binaries for Ubuntu from Alexander Pozdnyakov.

Problem is fixed in that repository now, new binaries were built from commit 7afa05a03ed87a95a46da798fd11cae60f390441

[scardine]

Please ignore the previous update, the package tesseract-ocr-por is not fixed yet.

I had to download from https://github.com/tesseract-ocr/tessdata/blob/master/best/por.traineddata and copy to /usr/share/tesseract-ocr/4.00/tessdata/por.traineddata

[amitdo]

You can also put the traineddata in a sub directory (like 'best') under your tessdata dir and use -l subdir/lang

[paulaceccon]

I got the "new" files from here, but I still get the same result.

[scardine]

@paulaceccon where did you copy them to and how are you calling tesseract? (in order to help you diagnose the problem we must be able to reproduce it)