Japanese just doesn't work, pretty much at all. #1436
Description
Hey so I'm looking for an OCR for linux and I found this but as expected it's just a tesseract wrapper. FWIW, tesseract is consistently inaccurate and wrong. Even on black on white clear high res text the kanjis (and hiragana) are wrong close to 80% to 90% of the times. I think the model or the library itself was just not made for japanese. Do you plan to change it to some other backend? Google lens works waaaay better (99% accuracy), but it's an online API, idk if you can find an alternative, I'm not a huge coder. I'm currently exploring EasyOCR but even then it's not great, or not as good as google lens... Thanks