search

dmMaze / BallonsTranslator

深度学习辅助漫画翻译工具, 支持一键机翻和简单的图像/文本编辑 | Yet another computer-aided comic/manga translation tool powered by deeplearning
GNU General Public License v3.0
2.48k stars 172 forks source link

Google OCR 有點問題 #544

Open Raiter123 opened 3 weeks ago

Raiter123 commented 3 weeks ago

Google OCR 有點問題 掃完OCR后發現每個原圖上斷句在譯文上變成空格了 有方法解決嗎 image image 還有啊OCR效果太强了連非常小的注音也掃進去了

bropines commented 3 weeks ago

Google OCR 有點問題 掃完OCR后發現每個原圖上斷句在譯文上變成空格了 有方法解決嗎 image image 還有啊OCR效果太强了連非常小的注音也掃進去了

Well, we can't edit what Google gives us. We only get strings. I can release a fix that can fix problems with text sticking. And I also don't understand Chinese, I'll see what I can do.

Raiter123 commented 3 weeks ago

s us. We only get strings. I can release a fix that can fix problems with text sticking. And I also don't understand Chinese, I'll see what I can do.

the word ocr by google ocr will a space when the original word at picture have Sentence

bropines commented 3 weeks ago

s us. We only get strings. I can release a fix that can fix problems with text sticking. And I also don't understand Chinese, I'll see what I can do.

the word ocr by google ocr will a space when the original word at picture have Sentence

I can throw in a good code, in theory it will help you, but I'm not sure

bropines commented 3 weeks ago

s us. We only get strings. I can release a fix that can fix problems with text sticking. And I also don't understand Chinese, I'll see what I can do.

the word ocr by google ocr will a space when the original word at picture have Sentence

I now understand what you mean (I just came back from vacation). The problem lies in the very specifics of GOOGLE recognition. We can't influence how he understands the text. Sometimes Japanese recognizes Vertical text as horizontal, which is fundamentally wrong. It's the same with spaces, Google can and does take these spaces into account when outputting to the UI on the site, but at the stage of receiving the text we can't always get what Google transmits to the UI on the site (he showed oleg, and wrote to us in the server's response ole g!). I'll sit on the fix, but mostly it's LensOCR crutches.

bropines commented 3 weeks ago

And about the small text. The same problem. Lens recognizes text wherever it sees it. Sometimes he doesn't take it into account. It all depends on the size of the image that was transmitted to him. Try to enlarge the text block for recognition, and maybe the result will be different

Raiter123 commented 2 weeks ago

And about the small text. The same problem. Lens recognizes text wherever it sees it. Sometimes he doesn't take it into account. It all depends on the size of the image that was transmitted to him. Try to enlarge the text block for recognition, and maybe the result will be different关于小文本。同样的问题。Lens 可以在任何看到的地方识别文本。有时他不会考虑到这一点。这完全取决于传输给他的图像的大小。尝试放大文本块进行识别，也许结果会有所不同

OK i try now