EliteOCR – Optical Character Recognition for The Commodities Market

[seeebek]

I just wish it was possible to run tesseract directly in the browser without installing anything on my PC or upload to a server... (like ocrad that was was compiled to pure javascript http://antimatter15.com/ocrad.js/demo.html)

I'm sure it is possible, just nobody did it yet.

Working mostly great. OCR has a bit of trouble with distinguishing 3, 6 and 8. I suppose that's just the nature of the beast with the in-game font making it difficult.

Awesome tool.

Yeah, 3,6 and 8 are the biggest nemesis of tesseract. It gets them often wrong even on other fonts. I'm slowly trying to compile tesseract 3.03. It supposedly improved this case at least a bit.

[Snuble]

I'm sure it is possible, just nobody did it yet.


Yeah, 3,6 and 8 are the biggest nemesis of tesseract. It gets them often wrong even on other fonts. I'm slowly trying to compile tesseract 3.03. It supposedly improved this case at least a bit.

I honestly have no time to start learning how to use Emscripten. It would be a couple of hours for me to deliver in-browser images to an in-browser script (if you've followed the crows source coordinate thread, you've already seen a couple of work in progress there). Neither the training data or a 300KB js script is anything to cause panic. Both can be cached easily with resulting 1 server download when something is updated. A simple online form to spot ocr-errors and completely in browser and client side would be perfect.

[zxctypo]

If you want online, live error correction, then you'd want to run ML server side, with learning from client side. I've forked the project, and a ML implementation for the numbers is in progress :-)

[seeebek]

What is MI?

[jayce1996]

Unfortunately this crashes on my machine with a MemoryError. I believe this is due to my screen size being 5040x1050. Is there a simple way to crop the middle 1/3 of the image (horizontally) as that's where the bit's we'd be interested in are? (I'm ok with python, but I don't know tesseract or numpy at all, and the docs are plain confusing)

[Inhumierer]

Excellent, commander!!! The OCR is working quite well, I couldn't make tesseract work like this. I tried, but I'm too stupid. When it gets to collecting the CSV in a database, I'm with you.

[seeebek]

Unfortunately this crashes on my machine with a MemoryError. I believe this is due to my screen size being 5040x1050. Is there a simple way to crop the middle 1/3 of the image (horizontally) as that's where the bit's we'd be interested in are? (I'm ok with python, but I don't know tesseract or numpy at all, and the docs are plain confusing)

Hi, good point. I will not provide it immediately but I will create the possibility in calibrate to crop very wide screenshots in the next version. Probably in two days...

Edit:
if interested I will do the crop with openCV or numpy. Imagemagick would add another library and increase the size of the tool.

[Inhumierer]

Unfortunately this crashes on my machine with a MemoryError. I believe this is due to my screen size being 5040x1050. Is there a simple way to crop the middle 1/3 of the image (horizontally) as that's where the bit's we'd be interested in are? (I'm ok with python, but I don't know tesseract or numpy at all, and the docs are plain confusing)

Yeah, easy. Let's swap our displays
No, honest: ImageMagick does the trick, look at http://www.imagemagick.org/index.php. There is a Win executable, too: http://www.imagemagick.org/script/bi...es.php#windows
You need the "crop" feature: http://www.imagemagick.org/Usage/crop/#crop

[jayce1996]

Oh just as a heads up to those who are thinking of using this excellent project as the basis for automating updates of a database. Beware, station names are not unique (I've seen two Forrester Ports in different star systems!)

- - - - - Additional Content Posted / Auto Merge - - - - -

Yeah, easy. Let's swap our displays
No, honest: ImageMagick does the trick, look at http://www.imagemagick.org/index.php. There is a Win executable, too: http://www.imagemagick.org/script/bi...es.php#windows
You need the "crop" feature: http://www.imagemagick.org/Usage/crop/#crop

Sure, I get that, I could use MSpaint to manually crop stuff before feeding it in, but it would have to be re-calibrated every time due to human error in the cropping, I was thinking more what seeebek suggested, something inside the app itself to automatically crop in RAM. I'd make the amend myself, only I can't work out how to work with the graphics libraries being used.

[zxctypo]

ML is machine learning. I should be submitting a pull request sometime later today with my changes :-)

What is MI?

[Cthulhus]

Excellent job ! thanks !

[seeebek]

Oh just as a heads up to those who are thinking of using this excellent project as the basis for automating updates of a database. Beware, station names are not unique (I've seen two Forrester Ports in different star systems!)

Yeah, in the next version you will have the option to get system name from logfiles and add to the table. So problem solved (soon)

[Megachill]

I am already doing that tho - since the log has the systemname - Checking system names with station names - these are IDs / keys in a table that are counter looked up. Problem would be if Frontier moved this out of the logfile...

They keep talking about an API (at least Michael is) so one can hope...

[DRY411S]

Well it would appear so. Unfortunately Trade Dangerous requires Python 3.4.1 and I have to use python 2.7 because of openCV. I could try to compile the newest source but it would cost me too much time.
Anyway in the future: maybe.

Is it possible to run both versions of python on the same machine?

[the-monk]

Such an awesome tool. Might have to give it a try it looks really useful.