skotz / captcha-breaking-library

Neural network, contour analysis, bitmap vector subtraction CAPTCHA solving library and scripting language with perceptive color space segmentation
GNU General Public License v3.0
81 stars 25 forks source link

Please support to provide document on how to train/generate db for solver! #9

Closed vikbeo closed 6 years ago

vikbeo commented 6 years ago

Thank Skotz for your very good lib.

I'm working on an captcha which I can work on the filter to have clean final image but currently I don't see how to train the solver and get db?

Actually I have just read previous Issue ticket but not sure how to implement the training properly.

Such as we need to change script to Train & Save instead of Solve? TRAIN, "I:\qqq" SAVE, I:\qqq\new.db"

I do see we need a test pool of as big as possible images but how can tell Solver which is good which is bad?

On your script in previous Issue: TRAIN, "qqq" FULLTEST, "qqq", "test.txt"

I even see the result: TOTAL CORRECT: 213 (213/266 = 80.08%)

Is that we are manually calculate or the lib support that?

Thank for your support in advanced.

My script. SetMode, all SetupSegmenter, BLOB, 30, 40, 4 SetupSolver, BVS, "ABCDEFGHIJKLMNOPQRSTUVWXYZ", 50, 60

DefinePreconditions Resize, 400, 140 RemoveSmallBlobs, 10, 4, 4 BlackAndWhite Grow Grow Median EndPreconditions

Solve, %IMAGE%

skotz commented 6 years ago

I just put together some new documentation to walk you through how to train a model. Please let me know if any parts are confusing and I'll try to update it accordingly.

Check out the solver walkthrough.

Thanks!

skotz commented 6 years ago

And here's a link to the folder with a working example and training scripts.

https://github.com/skotz/captcha-breaking-library/tree/master/Examples/Color%20CAPTCHA

vikbeo commented 6 years ago

Great. The route is much clearer! Thank a lot for your helpful lib and kind support.