Open juanj opened 3 years ago
Adding a preprocessing step can really improve the results.
Right now, the tesseract thresholding algorithm some times eats all the strokes of a kanji, leaving only the shape
From
To
It may be worth to use a different thresholding algorithm and let the user tweak it.
Removing furigana, speech bubble border and anything on the background gives results without junk
Adding a preprocessing step can really improve the results.
Right now, the tesseract thresholding algorithm some times eats all the strokes of a kanji, leaving only the shape
From
To
It may be worth to use a different thresholding algorithm and let the user tweak it.
Removing furigana, speech bubble border and anything on the background gives results without junk