Greek alphabets like Θ can't be recognized

[ttbuffey]

Environment

• Tesseract Version: 4.0
• Commit Number:
• Platform: mac

Current Behavior:

Command:
tesseract ~/Desktop/detector_sample_data-1.jpg ~/Desktop/result -l eng+ell+grc

Expected Behavior:

Greece National ID Card: Θ-420001 which has greek alphabets should be recognized, but 0-420001 is recognized

Suggested Fix:

for a english document which has few greek alphabets, the greek alphabets αβγδεζηθικλμνξοπρστυφχψωΑΒΓΔΕΖΗΘΙΚΛΜΝΞΟΠΡΣΤΥΦΧΨΩABEZHIKMNOPTYX should be recognized in the right way

[stweil]

This is not a software issue but more or less normal for OCR.

You can try using -c load_number_dawg=0 and see whether that helps.

Please use the Tesseract user forum for more questions.

[amitdo]

With the lstm engine, letters sequences that were not learnt during training have a low chance to be recognized.

[stweil]

Yes, training some Greece National ID Card (~100 or more) with tesstrain would help to get better results.

[ttbuffey]

I have a question same as described in the below issue
Shreeshrii/tess5train-fonts#15

@stweil @amitdo Could you please point out why the performance decrease after I fine tune the english model.