bertsky
commented
4 years ago @nai-kon Could you please specify in which PSM (page segmentation mode, see tesseract --help) you have been running? Also, a copy of the image you are describing would be helpful.
But with legacy dictionary, it seems done by full-match.
It shouldn't. So yes, what you are describing looks like a bug in the legacy engine. (And both engines should behave the same in that respect.)
What I afraid is should add the countless combinations of compound noun to dictionary.
You don't have to. Even for languages like German with whitespace-delimited words but undelimited compounds, you don't have to do that. (The dictionary will be applied at word boundaries with or without spaces.)
astrung
I'm using Tesseract with legacy dictionary (no lstm dictionary) for Japanese recognition. I added custom words on user-words dictionary, but it seems works when full-matched a word.
Japanese language has no spacing between words, like Germany. e.x. "トマトサラダ" (tomatosarada => tomato sarada)
When I put the image of "トマトサラダ" (tomatosarada) without dictionary, I got an incorrect result "トマトサうダ" (tomatosauda). So, I added "サラダ" (sarada) to user dictionary, but the result didn't changed. Next, I added "トマト" (tomato) and "サラダ" (sarada) individually to user dictionary, but the result didn't changed. Finally added the full word, "トマトサラダ" (tomatodarada), the result became correct.
I thought that dictionary correction was done by partial match, so I expected the words to add the dictionary is OK with each word of a compound noun. (e.x. add the "tomato" and "sarada", indivisually)
But with legacy dictionary, it seems done by full-match. What I afraid is should add the countless combinations of compound noun to dictionary.
Also, I've tried with lstm dictionary. "トマトサラダ" without dictionary became "ト マ ト サ ラ 人 ダ". So, I added "サラダ" (sarada) to user dictionary, the result became correct. With lstm dictionary, the dictionary correction seems done by partial match.
Now, I have two questions.
Q1. Does the Tesseract 3.X legacy engine correction done by full-match? If so, should I add the possible combination of compound noun to dictionary?
Q2. Does the Tesseract 4.X lstm engine correction done by partial match?
I searched around and couldn't find a solution. Please help me out. Thank you.
Environment