[Python] How to prioritize French OCR over Arabic when using Tesseract (fra+ara) on bilingual...

I'm working on scanned documents (registers) that contain both French and Arabic text.

When I run Tesseract OCR with lang='fra', all the French text is extracted perfectly.

But when I use lang='ara+fra' to handle both languages in one pass, I start getting weird errors:

• French words are misread or replaced (e.g., SOCIETE becomes 50012175)
• Some French company names become random Arabic-like characters (e.g., ALPHA becomes حمناطاحم)
• Arabic words work fine, but French gets corrupted.

What I want:


I’d like Tesseract to prioritize French whenever the text looks French (Latin letters), and only fall back to Arabic if it’s actually Arabic text.

I tried doing OCR in two passes (fra and ara) and combining them manually, but it’s hard to align words correctly since Tesseract splits and orders text differently for Arabic and French.

Question:


How can I ensure French text is not misinterpreted as Arabic when using Tesseract on bilingual documents?

Is there a way to:

• Prioritize fra over ara during a single OCR run?
• Or post-process the result from lang='ara+fra' to correct misclassified French text?
• Or a smarter way to combine fra and ara outputs?

Any tips, workarounds, or best practices would be appreciated


Tools:

• Tesseract v5
• pytesseract in Python
• OCR input: scanned PDFs/images (high quality) sample image

Continue reading...