Poker-AI.org
https://poker-ai.org/phpbb/

At last OCR has been defeated.
https://poker-ai.org/phpbb/viewtopic.php?f=26&t=2636		 Page 2 of 3
Author:  spears [ Tue Nov 26, 2013 9:34 am ]
Post subject:  Re: At last OCR has been defeated.
nefton wrote:
Sory for my english.


Absolutely no problem. I understand you fine. Your English is better than my Russian. Good posts
Author:  eugen [ Wed Nov 27, 2013 9:44 am ]
Post subject:  Re: At last OCR has been defeated.
nefton wrote:
Sory for my english.
I meant that I had won OCR. But now I understand how far this victory yet. Fixity table size, low noise immunity, and an application for screenshots. Still have to hope for incompetence of poker room. This is not a victory. I want to make very external OCR with the camera and the other computer. Unfortunately the time I have not so much, and I have to postpone the decision of external OCR.

Have you considered using video splitters instead?
Author:  nefton [ Wed Nov 27, 2013 10:45 am ]
Post subject:  Re: At last OCR has been defeated.
eugen wrote:
Have you considered using video splitters instead?


Yes, of course I considered this option.
Its only advantage - accurate transmission pixels and their colors (but do not forget that the first thing that will make the poker room for detecting robot - little change the position of different elements and their color)

Searching on the internet, I have not found anything suitable. You have to buy and try. Besides, the big question is whether it will suit you.
And the camera lie near :) Besides OCR = optical character recognition :)

p.s. Just looked at ebay. Yes, it seems a good solution. I will try it. (but with a camera of course mutch more beautiful :) )
Author:  eugen [ Wed Nov 27, 2013 12:16 pm ]
Post subject:  Re: At last OCR has been defeated.
The issue I have with a camera is that it will never be as precise as a split image. Which imo is very important, considering that poker rooms check for bots by slightly manipulating positioning of elements on the table.

But you're ahead of me on this, so I can only speak from theory.. :)
Author:  HontoNiBaka [ Sat Nov 30, 2013 12:22 am ]
Post subject:  Re: At last OCR has been defeated.
Can your scraper detect action sequences? If yes, how do you do it? I thought about just scraping the tables every 10 ms or so, but I am still afraid to miss some actions.
Author:  Gnomikian [ Sat Nov 30, 2013 12:36 am ]
Post subject:  Re: At last OCR has been defeated.
Quote:
Can your scraper detect action sequences? If yes, how do you do it? I thought about just scraping the tables every 10 ms or so, but I am still afraid to miss some actions.


Just analyse the table when it's your turn to play, that way you can get every action from your previous activity by deduction. No need to scrape the table every few ms imo.
Author:  nefton [ Sat Nov 30, 2013 12:40 am ]
Post subject:  Re: At last OCR has been defeated.
10 ms :shock: its too fast. 500ms more than enough. I wroute a class Hand_processor. He writes all regognized table info (from recognizer) to an array. Also he includes class FSM. (poker finite state machine). And when new info comes - he try to "clock" fsm. Based on all previous table infos.
It should be noted that I have not yet finished prototype of Hand_processor.
Author:  HontoNiBaka [ Sun Dec 01, 2013 11:54 pm ]
Post subject:  Re: At last OCR has been defeated.
How would you make your scraper more robust against noise and Stars slightly changing the colors etc.?

Atm I use a very simple algorithm, I just calculate the average difference between the screenshot and the training images, it works 100% if there is no noise. I also look for certain pixels to find the window etc. I use a tolerance of 5 for the RGB values, but that is easy to beat by stars.
Author:  nefton [ Mon Dec 02, 2013 12:14 am ]
Post subject:  Re: At last OCR has been defeated.
We must look a particular problem, to find the optimal solution. Against random noise while character recognition - convolution works well. To filter the range of colors (such as purple Call in ps) suitable conversion to HSV. (except red. was thinking to write own conversion).
Also would be nice to add to the recognizer detector confusing situations. That saved a screenshot.
Author:  shalako [ Mon Dec 02, 2013 1:48 am ]
Post subject:  Re: At last OCR has been defeated.
I have never seen an issue with PS changing pixel locations or color. Is this something new or ? I just checked it and everything works fine
Author:  HontoNiBaka [ Mon Dec 02, 2013 3:39 am ]
Post subject:  Re: At last OCR has been defeated.
I've read, that it's something they do, when they suspect you of botting. A human won't even notice it, but the bot might crash. You are probably not suspicious in their eyes.
Author:  HontoNiBaka [ Thu Dec 05, 2013 12:26 am ]
Post subject:  Re: At last OCR has been defeated.
Image

Your thread gave me motivation to work on my scraper again. So far it works 100%, I feel, like I have discovered the secret to good scraping. I don't know, how robust it is though.
Author:  nefton [ Thu Dec 05, 2013 1:18 am ]
Post subject:  Re: At last OCR has been defeated.
My congratulations! *drink*
Good work.
First word(hand number) I recognized about a month.
And last (table pot) took me less an hour (with collect chars to comparison)

That means that ocr has ben defeted :) Not mater how the difficult of this problem will increase (various table size, change pixel color, external vga grabber, or even webcam :).

Congards again.
Author:  shalako [ Thu Dec 05, 2013 9:15 pm ]
Post subject:  Re: At last OCR has been defeated.
Quote:
I've read, that it's something they do, when they suspect you of botting. A human won't even notice it, but the bot might crash. You are probably not suspicious in their eyes.


Well if your already flagged then OCR will be the least of your problems.
Author:  spears [ Thu Dec 05, 2013 10:01 pm ]
Post subject:  Re: At last OCR has been defeated.
https://poker-ai.org/archive/www.pokerai ... =79&t=3941
Author:  HontoNiBaka [ Fri Dec 06, 2013 5:43 am ]
Post subject:  Re: At last OCR has been defeated.
Now I can also scrape stats:

Image

I deleted the stacksizes from the image.

I think, now I pretty much have all the functionality I will need, time for code cleanups, testing and debugging.
Author:  nefton [ Fri Dec 06, 2013 5:47 am ]
Post subject:  Re: At last OCR has been defeated.
stats from poker tracker?
Author:  HontoNiBaka [ Fri Dec 06, 2013 6:39 am ]
Post subject:  Re: At last OCR has been defeated.
Yes, PokerTracker 4 stats, scraped directly from the VM, so no SQL or anything.

However, I don't know, if that is the best way. It is secure, but I can only use a few stats. I might look into using SQL on my host and getting stats via SQL queries.

If anyone has tipps, how to do that securelly, I would be thankful.
Author:  nefton [ Fri Dec 06, 2013 7:20 am ]
Post subject:  Re: At last OCR has been defeated.
think you should try to write the bot. Many questions like "secure scraping poker tracker stats" will disappear by themselves.
Author:  MrNice [ Fri Dec 06, 2013 7:52 am ]
Post subject:  Re: At last OCR has been defeated.
Hi HontoNiBaka,

Congrats!

Which techniques are you using for OCR ? tesseract?

Regards,

Mrnice
Page 2 of 3 All times are UTC
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/