[X] My Assistant

Yo!

I wanna get an idea of what OCR programs I can use. I translate by using Google Translate for a base idea for meaning, then double check everything with wiktionary, tanoshii, and searches for Japanese slang, plus going over context within images or paragraphs to make sure I'm not too off the mark. I've been translating a lot of Moonrevenge and Milda7 because their work includes proper, copyable text outside the image in most cases (and a few Dining sets thanks to blamekilly pasting the text in copyable format in the comments).

However, there's a lot more I want to try and translate that only has the text within the image. I can copy text with KanjiTomo from more plain backgrounds, like black text on flat white backgrounds, but things get more complicated in CG sets. On top of that, fonts confuse the program a little, and if the text has a thick outline or is over a particularly busy background, the program is effectively useless. It can also only pick up four characters at a time, which is a big problem with text heavy sets.

TL;DR: Are there any better OCRs than KanjiTomo for text on busy backgrounds or with thick outlines? Any that can get more than four characters at a time?

You can try Capture2Text (last time I used it it wasn't good tho)

But Google Translate sucks (in fact all machine translation sucks), better to learn Japanese yourself.

This post has been edited by N04h: Jul 7 2018, 00:53

You can try Capture2Text (last time I used it it wasn't good tho)

But Google Translate sucks (in fact all machine translation sucks), better to learn Japanese yourself.

Oh no question, I'm actually trying to learn with all the double checking (which is how I know it's never a good idea to trust the bloody thing).

I've tried Capture2Text as well and man it's a disaster.

I don't know a OCR software besides Kanjitomo, which thankfully works 80% of the times, but when it isn't working I usually draw the characters on my phone, which is still nicer than looking them up in the dictionary by stroke count and components. I use (but cannot really recommend) Kanji Draw on Android, but I guess there is better software out there. And when the resolution is so bad that I cannot redraw them, then OCR cannot read them either.

I dont think most programs can do a 100% capture without some problems.

If you ask me, for a free software, Capture2Text does a much better job now than 1½ years ago, but you will still have to accept that some letters will be a hit or miss, especially with や/ゃ and つ/っ.

When it comes to typing Japanese IME, I actually think Google is a very good tool. Google (ignoring 99% of it's own translations), Firefox/Chrome with Rikaichan is quite nice (IMG:[invalid] style_emoticons/default/tongue.gif)
But yea, actually learning Japanese would be better.

If you are on windows 10, I'd recommend using Takaboto as well. To find kanji based on the components used, or if you know the romaji for a kanji, you can type that in to find it too.

This post has been edited by q91: Jul 17 2018, 19:16

I don't think OCR is a necessity for translating manga. You'll just type the translations line by line in a notepad, so why would you need the JP transcripts?

Even if you prefer/need having the Japanese transcripts (like I do for archiving/quick rechecks), you're better off typing them using a keyboard since it's faster that way. It also reinforces your comprehension of the story, since you read, type, then read it again. You might think that OCR is faster by itself, but typing the JP lines doesn't really take that much time.

Machine translation is generally a bad idea too IMO. You might think that you can fix the machine TL dialogues into intelligible sentences but translating languages isn't that simple, especially between non-related languages like JP->EN. You might think that you can guess and paraphrase machine TL dialogues of any stories out there, but rewrites and wrong translations will always be caught since people will still try to check the raws in time.