ROM Hack Nftr based ocr?

[rastsan]

Ocr- optical character recognition.
I was wondering if any of you know of one?
Or would be willing to make one?

Nuance omnipage professional is giving me way to much trouble even reading the image (taken directly from the nftr) to help make a training file (which would then be used to read the screenshots).
14 tries so far - the output doesn't match the image does not match any characters in the nftr itself, or on the screen. it for some reason always displays in 114 or 145 size font. No matter how many times I change the settings to 11 point.
so after spending way to much time on this....
I thought I would ask.

If there isn't any, would someone please make one?

I believe this would be very useful to the community....
Pretty please...

 

[habababa]

try converting your nftr to a 1bpp image and have kanjiocr scan it.

 

[FAST6191]

So if I understand it you have a rom, it has an NFTR or some other known font which you took a screenshot of to try to train an OCR program which you then plan to use on game screenshots (I am guessing some hardsubbed video or something)


Crystaltile2 has a measure of OCR in it (tools menu in the graphics viewer section) and better yet it can seemingly do it for anything (although in my test just now I was having a bit of trouble with 2 tile characters). More so than with most OCR programs it looks like you will have to prod it along.

Failing that I would cheat- kick your image through something like avisynth and make it a video.
Even 6 years ago I saw various video OCR programs tear through whatever the hardsub anime crowd were using on moving backgrounds although looking around aside from some DVD stuff (subtitle creator and subtitle edit) nobody took that much further and it is still subrip http://zuggy.wz.cz/guides/video.htm
I have toyed with some of the static image stuff but been unimpressed.

As for making one OCR is traditionally one of the hard problems faced by computers and this goes double for the Asian languages if you are heading there. Some stuff has gone on in recent years (see the anti bot methods).

 

[rastsan]

Well thanks guys but not really what I meant.
I have used readiris and other versions of their software (for work some years ago).
No what I was hoping was to inspire someone to build one...

@ fast eh I have tried ct2's inbuilt ocr believe me easier to use omnipage... it focuses on just those three sizes of fonts nothing else. If I could have a train function, might be better.
edit thanks for pointing me to subrip

I finally got omnipage to let me train the characters. Omnipage though is surprisingly having trouble with even the english in this font.
Now I already have the table file and the game is nftr based I just thought "Why am I picking and pecking and searching so slowly for the script to organize it when I can just ocr the screen shots and output to that same text file, oh and make it easier use the nftr itself to read the screenshot..."

the alternative to this is to ocr the nftr graphic itself and then have a training file in the ocr that can then be used on the screenshots.
Build up your script quick and easy without the messy pick and peck and search in between....(for those projects that have the script in a seemingly disorganized all over way)
I was trying to say use the nftr itself as the training file to Read the screenshots.... as there must be some way to do that...